.editorconfig

@@ -0,0 +1,13 @@
+# Root editor/formatter config. end_of_line=lf makes dotnet-format agree with
+# .gitattributes (eol=lf) — without this, format-on-Windows wants CRLF while git
+# stores LF, and the pre-commit/CI format gates flip-flop forever.
+root = true
+
+[*]
+end_of_line = lf
+insert_final_newline = true
+charset = utf-8
+
+[*.cs]
+indent_style = space
+indent_size = 4

.env.example

@@ -15,3 +15,6 @@ FRONTEND_ORIGIN=http://localhost:8081
 # Set DEV_MODE=true and MAX_MESSAGES=1000 to test against a large mailbox.
 DEV_MODE=false
 MAX_MESSAGES=0
+
+# Nightly DB backup rotation (days of dumps to keep in ./backups)
+BACKUP_KEEP_DAYS=7

.env.staging.example

@@ -0,0 +1,17 @@
+# Copy to deploy/.env.staging and fill in. Used by:
+#   docker compose -p inboxintel-staging --env-file deploy/.env.staging \
+#     -f docker-compose.yml -f docker-compose.staging.yml up
+# (or ./deploy/up.ps1 -Staging). Kept separate from deploy/.env (production) so
+# staging can never touch prod credentials, DB, or Google project.
+POSTGRES_PASSWORD=change-me-staging
+
+# Use a SEPARATE Google OAuth client for staging with redirect URI:
+#   http://localhost:18081/signin-google
+GOOGLE_CLIENT_ID=
+GOOGLE_CLIENT_SECRET=
+
+AI_MODE=Disabled
+FRONTEND_ORIGIN=http://localhost:18081
+
+# Staging caps the initial mailbox sync so rehearsals are fast.
+MAX_MESSAGES=2000

.gitea/workflows/ci.yml

@@ -14,7 +14,7 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-dotnet@v4
         with:
-          dotnet-version: '8.0.x'
+          dotnet-version: '10.0.x'
       - name: Restore
         run: dotnet restore InboxIntel.sln
       - name: Build
@@ -31,8 +31,50 @@ jobs:
       - uses: actions/checkout@v4
       - uses: actions/setup-node@v4
         with:
-          node-version: '20'
+          node-version: '22'
       - name: Install
         run: npm ci
       - name: Build
         run: npm run build
+
+  # AUDIT L-10: the pre-commit hook enforces formatting locally, but --no-verify or web edits
+  # can bypass it — this makes the same check a server-side gate.
+  format:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-dotnet@v4
+        with:
+          dotnet-version: '10.0.x'
+      - name: dotnet format (verify only)
+        run: dotnet format InboxIntel.sln --verify-no-changes
+
+  # AUDIT M-7: the search paths the InMemory provider can't translate (FTS ranking,
+  # ts_headline, pg_trgm, pgvector) previously had only manual verification. This job runs
+  # the Category=LiveDb tests against a real pgvector Postgres service container.
+  db-tests:
+    runs-on: ubuntu-latest
+    services:
+      postgres:
+        image: pgvector/pgvector:pg16
+        env:
+          POSTGRES_USER: test
+          POSTGRES_PASSWORD: test
+          POSTGRES_DB: test
+    env:
+      LIVEDB_CONNECTION: "Host=postgres;Port=5432;Database=test;Username=test;Password=test"
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-dotnet@v4
+        with:
+          dotnet-version: '10.0.x'
+      - name: Wait for Postgres
+        run: |
+          for i in $(seq 1 30); do
+            (echo > /dev/tcp/postgres/5432) 2>/dev/null && exit 0
+            sleep 1
+          done
+          echo "Postgres service did not become reachable" >&2
+          exit 1
+      - name: Live-DB tests
+        run: dotnet test InboxIntel.sln --filter "Category=LiveDb" --verbosity normal

.gitea/workflows/deploy-prod.yml

@@ -0,0 +1,45 @@
+name: Deploy Production
+
+# Production promotion. The APPROVAL GATE is the git tag: production only ever
+# deploys a tagged release cut on main (see docs/WORKFLOW.md §5). Cutting the tag
+# is the deliberate, auditable "approve to go live" action — and the tag doubles
+# as the rollback target. workflow_dispatch adds a manual "Run workflow" button
+# for re-deploys/rollbacks.
+#
+# STATUS: inactive until (a) the Linux production server exists and (b) a
+# self-hosted Gitea runner is registered on it with labels [self-hosted, production].
+# Tag pushes before then will queue harmlessly. This file is the wiring, ready to
+# switch on — review the deploy step for your server before first use.
+on:
+  push:
+    tags: ['v*']
+  workflow_dispatch:
+    inputs:
+      ref:
+        description: 'Tag or commit to deploy (e.g. v1.2.0)'
+        required: true
+
+jobs:
+  deploy:
+    runs-on: [self-hosted, production]
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          # Deploy the exact tag that triggered the run (immutable), or the
+          # ref given to a manual dispatch.
+          ref: ${{ github.event.inputs.ref || github.ref_name }}
+          fetch-depth: 0
+
+      - name: Deploy release to production
+        run: |
+          echo "Deploying ${{ github.event.inputs.ref || github.ref_name }} to production"
+          # deploy/.env lives on the server (never in git). up.sh validates it,
+          # builds the Linux images, applies EF migrations on boot, and starts
+          # the stack behind the nginx reverse proxy.
+          ./deploy/up.sh --proxy
+
+      - name: Smoke check
+        run: |
+          sleep 5
+          curl -fsS http://localhost/ >/dev/null && echo "Prod responding on :80" || \
+            { echo "Smoke check failed"; exit 1; }

.gitea/workflows/deploy-staging.yml

@@ -0,0 +1,70 @@
+name: Deploy Staging
+
+# Continuous deployment to the LOCAL staging stack. Fires when develop advances
+# (i.e. after a PR is merged into develop), rebuilding and restarting the isolated
+# staging stack on this machine.
+#
+# Runs on the self-hosted host-mode runner (labels: self-hosted, windows) so it can
+# reach the host's Docker and publish to localhost:18081. Because the runner does a
+# fresh checkout that (correctly) does NOT contain the git-ignored deploy/.env.staging,
+# the staging secrets are supplied as Gitea Actions secrets and the env file is
+# regenerated here at deploy time.
+on:
+  push:
+    branches: [develop]
+  workflow_dispatch: {}      # also allow a manual "Run workflow" from the Gitea UI
+
+jobs:
+  deploy:
+    runs-on: [self-hosted, windows]
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Write staging env from secrets
+        shell: powershell
+        # Per-line writes (not a here-string) so the step can't be broken by how the
+        # runner indents/wraps the script. Gitea substitutes ${{ secrets.* }} first;
+        # ascii = no BOM, which docker compose's env parser needs.
+        run: |
+          Set-Content deploy/.env.staging "POSTGRES_PASSWORD=${{ secrets.STAGING_POSTGRES_PASSWORD }}" -Encoding ascii
+          Add-Content deploy/.env.staging "GOOGLE_CLIENT_ID=${{ secrets.STAGING_GOOGLE_CLIENT_ID }}" -Encoding ascii
+          Add-Content deploy/.env.staging "GOOGLE_CLIENT_SECRET=${{ secrets.STAGING_GOOGLE_CLIENT_SECRET }}" -Encoding ascii
+          Add-Content deploy/.env.staging "AI_MODE=Disabled" -Encoding ascii
+          Add-Content deploy/.env.staging "FRONTEND_ORIGIN=http://localhost:18081" -Encoding ascii
+          Add-Content deploy/.env.staging "MAX_MESSAGES=2000" -Encoding ascii
+          Write-Host "wrote $((Get-Content deploy/.env.staging).Count) env lines"
+
+      - name: Redeploy staging stack
+        shell: powershell
+        run: |
+          docker compose -p inboxintel-staging `
+            --env-file deploy/.env.staging `
+            -f docker-compose.yml -f docker-compose.staging.yml `
+            up -d --build
+          docker compose -p inboxintel-staging ps
+
+      - name: Verify staging health
+        shell: powershell
+        # 'up -d' returns as soon as containers START, so a container that crashes
+        # on boot (e.g. bad DB password) would still report success. Poll the actual
+        # endpoints and fail the job if either isn't serving, dumping api logs so the
+        # cause is visible in the run. ASCII only (Windows PowerShell reads .ps1 as ANSI).
+        run: |
+          $ok = $false
+          foreach ($i in 1..20) {
+            Start-Sleep -Seconds 3
+            try { Invoke-WebRequest "http://localhost:18081/" -UseBasicParsing -TimeoutSec 5 | Out-Null; $fe = 200 }
+            catch { $fe = 0 }
+            # A 401 from the api means it is serving (auth enforced); Invoke-WebRequest
+            # throws on non-2xx, so read the status code off the exception.
+            try { Invoke-WebRequest "http://localhost:18080/api/v1/auth/me" -UseBasicParsing -TimeoutSec 5 | Out-Null; $api = 200 }
+            catch { $api = $_.Exception.Response.StatusCode.value__; if (-not $api) { $api = 0 } }
+            Write-Host "attempt $i - frontend=$fe api=$api"
+            if ($fe -eq 200 -and $api -gt 0) { $ok = $true; break }
+          }
+          if (-not $ok) {
+            Write-Host "Staging health check FAILED. Last 40 api log lines:"
+            docker logs inboxintel-staging-api-1 --tail 40
+            exit 1
+          }
+          Write-Host "Staging healthy - Frontend http://localhost:18081  API http://localhost:18080"

.gitea/workflows/renovate.yml

@@ -0,0 +1,35 @@
+name: Renovate
+
+# RECOMMENDATIONS #2: automated dependency-update PRs (NuGet, npm, Dockerfiles, Actions)
+# that ride the existing required CI gates. Runs weekly + on demand.
+#
+# ONE-TIME SETUP (manual): create a Gitea personal access token with scopes
+# repo (rw) + user (r) + issue (rw) + organization (r), and add it as the Actions
+# secret RENOVATE_TOKEN (repo Settings -> Actions -> Secrets). Without the secret this
+# workflow fails fast with a clear message. See https://docs.renovatebot.com/modules/platform/gitea/
+on:
+  schedule:
+    - cron: '30 4 * * 1'      # Mondays 04:30 UTC
+  workflow_dispatch: {}
+
+jobs:
+  renovate:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Require RENOVATE_TOKEN
+        run: |
+          if [ -z "${{ secrets.RENOVATE_TOKEN }}" ]; then
+            echo "RENOVATE_TOKEN secret is not set — see the comment at the top of this workflow." >&2
+            exit 1
+          fi
+      - name: Run Renovate
+        uses: https://github.com/renovatebot/github-action@v40.3.6
+        with:
+          token: ${{ secrets.RENOVATE_TOKEN }}
+        env:
+          RENOVATE_PLATFORM: gitea
+          RENOVATE_ENDPOINT: https://git.cesnimda.uk/api/v1
+          RENOVATE_REPOSITORIES: cesnimda/Inboxintel
+          RENOVATE_ONBOARDING: "false"
+          RENOVATE_REQUIRE_CONFIG: optional
+          LOG_LEVEL: info

.gitea/workflows/security.yml

@@ -0,0 +1,72 @@
+name: Security
+
+# Scans run alongside CI on every PR and on pushes to the long-lived branches.
+# This is the DETECTIVE layer (backstop). The PREVENTIVE layer is the local
+# pre-commit hook — this catches anything that slipped past it (e.g. --no-verify)
+# and scans the full history, not just the staged diff.
+on:
+  push:
+    branches: [main, develop]
+  pull_request:
+
+jobs:
+  secrets:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0            # full history so gitleaks scans every commit
+      - name: Secret scan (gitleaks)
+        # Run the binary directly — the container-mode runner has no Docker socket,
+        # so `docker run` isn't available inside a job.
+        run: |
+          GITLEAKS_VERSION=8.18.4
+          curl -sSL "https://github.com/gitleaks/gitleaks/releases/download/v${GITLEAKS_VERSION}/gitleaks_${GITLEAKS_VERSION}_linux_x64.tar.gz" -o /tmp/gitleaks.tar.gz
+          tar -xzf /tmp/gitleaks.tar.gz -C /tmp gitleaks
+          /tmp/gitleaks detect --source=. --redact --verbose --exit-code=1
+
+  dependencies:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-dotnet@v4
+        with:
+          dotnet-version: '10.0.x'
+      - name: Restore
+        run: dotnet restore InboxIntel.sln
+      - name: .NET vulnerable packages (fail on any)
+        # Match dotnet's own "has the following vulnerable packages" line rather
+        # than raw severity words, so package/project names can't false-positive.
+        run: |
+          dotnet list InboxIntel.sln package --vulnerable --include-transitive 2>&1 | tee vuln.txt
+          if grep -q "has the following vulnerable" vuln.txt; then
+            echo "::error::Vulnerable NuGet packages detected — see the table above."
+            exit 1
+          fi
+          echo "No vulnerable NuGet packages."
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+      - name: npm audit (production deps, fail on high/critical)
+        # Only production dependencies ship to users; dev-only toolchain advisories
+        # (Vite/PostCSS/etc.) shouldn't block a merge.
+        working-directory: frontend
+        run: npm audit --omit=dev --audit-level=high
+
+  # RECOMMENDATIONS #9: SAST. Semgrep community rules for C#/JS + OWASP/secrets patterns —
+  # catches injection/crypto-misuse classes the other gates (gitleaks, dep-audit, tests)
+  # don't look for. Advisory at first (not a required check); promote once tuned.
+  sast:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      # The runner image lacks pip, and a semgrep job-container lacks the node that
+      # actions/checkout needs — so install pip via apt on the standard image.
+      - name: Install semgrep
+        run: |
+          sudo apt-get update -qq && sudo apt-get install -y -qq python3-pip pipx
+          pipx install semgrep
+      - name: Semgrep scan
+        run: |
+          export PATH="$HOME/.local/bin:$PATH"
+          semgrep scan --config p/csharp --config p/javascript --config p/security-audit             --exclude 'frontend/dist' --exclude '**/bin' --exclude '**/obj'             --error --quiet

.gitignore

@@ -21,6 +21,9 @@ frontend/.vite/
 appsettings.*.local.json
 secrets.json
 
+## DB backups (never commit dumps)
+backups/
+
 ## Logs
 logs/
 *.log
@@ -54,6 +57,7 @@ lpt[1-9].*
 *.code-workspace
 .env.*
 !.env.example
+!.env.staging.example
 .next/
 dist/
 build/
@@ -66,3 +70,6 @@ vendor/
 coverage/
 .cache/
 tmp/
+
+# agent worktrees (never commit)
+.claude/worktrees/

AUDIT_REPORT.md

@@ -0,0 +1,267 @@
+# InboxIntel — Full Application Audit Report
+
+**Date:** 2026-07-02 · **Scope:** entire repository (backend `src/`, `frontend/`, migrations,
+CI/CD `.gitea/workflows/`, `docker-compose*`, `deploy/`, git history) · **Phase:** 1 (read-only)
+
+**Stack:** .NET 8 / ASP.NET Core (Clean Architecture) · EF Core + PostgreSQL (pgvector image)
+· React 18 + Vite SPA · Google OAuth2 (cookie session) · Gitea Actions CI/CD · self-hosted
+Docker Compose (single-operator deployment).
+
+## Executive summary
+This codebase is **markedly more hardened than typical** — previous hardening phases (V-01…V-12,
+visible in code comments) already addressed the classic killers: IDOR (EF global query filters),
+SSRF (a genuinely thorough `SafeHttpGuard` with DNS-rebinding defence), token encryption at rest,
+non-root containers, stack-trace suppression, forwarded-header trust, secret hygiene (nothing in
+git history), and confirmed-destructive-actions. **No Critical findings.** The most significant
+issues are: **registered input validators that never execute (H-1)**, **no rate limiting (H-2)**,
+and a **data-at-rest posture** for email bodies that should be a deliberate, documented decision
+(H-3). CI has real security gates (gitleaks + dependency audit) already.
+
+Severity counts: **Critical 0 · High 3 · Medium 6 · Low 7**
+
+---
+
+## 1. Security
+
+### H-1 · FluentValidation validators are registered but NEVER executed
+- **File:** `src/InboxIntel.Application/DependencyInjection.cs:11` (registration) vs
+  `src/InboxIntel.Api/Program.cs` (no `AddFluentValidationAutoValidation()` anywhere; no
+  controller injects `IValidator<>`).
+- **Why it's a problem:** `Validators.cs` defines real safety rules — `CleanupRequestValidator`
+  requires `Confirmed=true` for Trash/HardDelete, `SearchRequestValidator` bounds paging, etc. —
+  but FluentValidation 11.x requires explicit auto-validation enablement, which is absent. The
+  rules are dead code; requests reach services unvalidated. *Mitigating factor:* the services
+  re-check the most dangerous rules (`CleanupService.cs:48` re-enforces `Confirmed`), and the
+  controller clamps paging — so this is defence-in-depth loss, not an open hole. But rules that
+  exist **only** in the validators (e.g. `From <= To`, LabelId-required) are silently unenforced.
+- **Fix:** add `services.AddFluentValidationAutoValidation()` (package already referenced) in
+  `Program.cs`; add a regression test posting an invalid DTO and asserting 400.
+- **Blast radius:** isolated (1 line + tests). Verify no existing client sends payloads that
+  would now 400.
+
+### H-2 · No rate limiting on any endpoint
+- **File:** `src/InboxIntel.Api/Program.cs` (no `AddRateLimiter`/middleware anywhere).
+- **Why:** login (`/auth/login` → OAuth), search (now with FTS + trigram + future AI), export
+  (PDF generation), unsubscribe (server-side outbound HTTP), and AI endpoints are all
+  unthrottled. A single authenticated user (or an unauthenticated client hammering
+  `/auth/login`/`/appinfo`) can exhaust CPU/DB/outbound quota. For the future multi-user platform
+  (see `docs/discovery/multi-provider/`), this is a prerequisite.
+- **Fix:** ASP.NET Core's built-in `RateLimiter` — a global fixed-window per-IP policy + stricter
+  policies on `auth`, `export`, `unsubscribe`, `ai`. Return 429 with Retry-After.
+- **Blast radius:** isolated (Program.cs + policy constants + tests).
+
+### Verified-good (no finding)
+- **Authorization / IDOR:** `[Authorize]` on `ApiControllerBase`; only `AppInfo` + `Auth.login`
+  are `[AllowAnonymous]` (correct). Global query filters on every tenant-scoped entity enforce
+  per-user isolation even if a query forgets its `Where` — covered by `TenantIsolationTests`.
+- **Injection:** all data access via EF parameterisation; no raw SQL string concatenation in app
+  code; FTS uses `websearch_to_tsquery` parameters. Frontend has **zero**
+  `dangerouslySetInnerHTML`/`innerHTML`/`eval`; the search-highlight feature deliberately uses
+  non-HTML sentinels rendered as escaped React elements (XSS-safe by construction).
+- **SSRF:** `SafeHttpGuard` validates scheme, resolves DNS and checks **every** address against
+  loopback/private/link-local/CGNAT/metadata/multicast (v4+v6, v4-mapped), fails closed on
+  unknown families, and is paired with `AllowAutoRedirect=false` (`DependencyInjection.cs:65`).
+  This is better than most production code.
+- **Secrets:** none hardcoded (empty placeholders in `appsettings.json`); no `.env` ever
+  committed (git history checked); real secrets live in git-ignored `deploy/.env*` and Gitea
+  Actions secrets; pre-commit hook + CI gitleaks scan both guard regressions.
+- **Session cookies:** HttpOnly, SameSite=Lax, Secure-always outside dev, 401-not-redirect for
+  XHR. Forwarded headers only trusted from configured proxy CIDRs (V-08).
+- **Headers:** nosniff, X-Frame-Options DENY, Referrer-Policy, COOP on API; HSTS in prod.
+
+### M-1 · No absolute session lifetime (sliding-only)
+- **File:** `Program.cs:58-59` — `ExpireTimeSpan = 7d` with `SlidingExpiration = true`.
+- **Why:** a session that's touched at least weekly renews forever; a stolen cookie never
+  expires as long as the attacker uses it. No server-side revocation list exists either
+  (cookie is self-contained).
+- **Fix:** add an absolute cap via an `issued-at` claim checked in `OnValidatePrincipal`
+  (e.g. re-auth after 30 days), and/or a session-stamp validated against the DB to enable
+  revocation. (The multi-provider design doc 02/06 already specs DB-backed sessions — this
+  aligns.)
+- **Blast radius:** isolated.
+
+### M-2 · No Content-Security-Policy on the SPA or API
+- **File:** `Program.cs:156-167` (comment says CSP "report-only for now" but none is actually
+  set); `frontend/nginx.conf` serves the SPA without security headers.
+- **Why:** CSP is the main mitigation layer against any future XSS slip; currently absent.
+- **Fix:** add CSP (default-src 'self'; connect-src API origin; etc.) + nosniff/XFO to the SPA
+  nginx config; optionally a report-only CSP on the API.
+- **Blast radius:** isolated (nginx conf + one middleware line), needs SPA smoke-test (Chart.js
+  inline styles etc.).
+
+### L-1 · CSRF: no antiforgery tokens (accepted-risk, documented here)
+- SameSite=Lax cookies + strict CORS allowlist + JSON-only POST bodies make classic CSRF
+  impractical in modern browsers. Acceptable for now; revisit if cookie SameSite is ever
+  relaxed or non-JSON form endpoints are added.
+
+### L-2 · `AllowedHosts: "*"` (`appsettings.json`)
+- Host-header filtering disabled; low risk behind the proxy but set it to the real hostnames
+  at production deployment.
+
+---
+
+## 2. Data protection (top priority)
+
+**Inventory of sensitive data:**
+| Data | Where | At-rest protection | Verdict |
+|---|---|---|---|
+| Google OAuth **refresh/access tokens** | `users.EncryptedRefreshToken` | **Encrypted** (Data Protection API, keys on `/keys` volume) | ✅ good |
+| **Passwords** | — | **None stored** (OAuth-only login; no password column exists) | ✅ best-possible |
+| **Email bodies / subjects / snippets** | `emails.BodyText` etc. | **Plaintext** in Postgres | ⚠️ **H-3** below |
+| Sender names/addresses (PII) | `senders`, `emails` | Plaintext | part of H-3 |
+| User email + display name | `users` | Plaintext | part of H-3 |
+| DB password | `deploy/.env` (git-ignored) + Actions secret | not in repo | ✅ |
+| Data Protection keys | `/keys` Docker volume | filesystem | M-3 below |
+| IPs / payment / health data | — | not collected | ✅ n/a |
+
+### H-3 · Email content stored in plaintext at rest, with unbounded retention
+- **File:** `src/InboxIntel.Domain/Entities/Email.cs` (`BodyText`, `Snippet`, `Subject`);
+  Postgres `pgdata` volume; also flows into PDF/CSV/JSON **exports** and (when enabled) to the
+  local Ollama process.
+- **Why:** the entire product is a copy of the user's mailbox. On this self-hosted,
+  single-operator deployment the DB lives on the operator's own disk — a defensible posture —
+  but: (a) anyone with disk/volume/backup access reads all mail; (b) there is **no retention or
+  purge policy** (mail persists even after unsubscribe/cleanup in Gmail; "remove account"
+  flows don't exist yet); (c) the planned **multi-user platform** (docs/discovery/multi-provider)
+  makes plaintext-bodies-readable-by-host-admin a real privacy issue (its own security doc
+  promises "admins never read users' mail" — the DB must back that up).
+- **Fix (phased):** 1) *Document* the current posture in README/threat model (deliberate,
+  local-first). 2) Enable **pgcrypto/field-level encryption or full-disk/volume encryption**
+  before any multi-user deployment. 3) Add a **data-retention setting** + purge job and an
+  account-deletion path (GDPR-style erasure). 4) Ensure DB **backups** inherit the same
+  protection.
+- **Blast radius:** documentation = trivial; field-level encryption = **large** (touches search
+  — FTS can't index encrypted columns; would need architectural decision). Recommend
+  volume-level encryption + retention/deletion first.
+
+### M-3 · Data Protection keys stored unencrypted on the `/keys` volume
+- **File:** `Program.cs:26-28` — `PersistKeysToFileSystem` without `ProtectKeysWith*`.
+- **Why:** whoever reads the volume can decrypt all stored refresh tokens. Same-disk-as-DB
+  caveat applies, but keys and ciphertext living side-by-side weakens the encryption's value.
+- **Fix:** `ProtectKeysWithCertificate(...)` (cert from env/secret), or OS-level DPAPI on
+  Windows hosts; at minimum document the volume-permissions requirement.
+- **Blast radius:** isolated, but requires a key-migration step for existing tokens.
+
+### M-4 · Default DB credentials in `appsettings.json`
+- **File:** `appsettings.json:3` — `Password=inboxintel` as the fallback connection string.
+- **Why:** if a deployment forgets the env override, the app happily connects with a guessable
+  password (compose enforces `POSTGRES_PASSWORD` but a non-compose deployment might not).
+- **Fix:** empty the default and fail fast at startup with a clear message when unset.
+- **Blast radius:** isolated (plus updating dev docs to use user-secrets).
+
+### Logging & transit — verified
+- **Logs:** no token/password logging found; the one PII-ish log is `SmtpEmailSender.cs:28`
+  (recipient address + subject at Info when SMTP is unconfigured) — acceptable, downgrade to
+  Debug if desired (L-3). Serilog request logging does not include query strings or bodies.
+- **Transit:** TLS terminates at nginx (HSTS enabled in prod); API+DB bound to loopback/compose
+  network only. **DB connection itself is non-TLS** — fine while Postgres is co-located on the
+  compose network; **must add `SSL Mode=Require` if the DB ever moves to another host** (L-4).
+
+---
+
+## 3. Dependencies & supply chain
+- **`dotnet list package --vulnerable --include-transitive`:** **clean** (all projects) — the
+  earlier round of CVE pins (System.Text.Json 8.0.6 etc.) is holding, and CI re-checks on every PR.
+- **`npm audit` (production deps):** **0 vulnerabilities**.
+- **M-5 · `npm audit` (dev deps):** vite 5.x depends on a vulnerable **esbuild** (1 moderate,
+  1 high, dev-server-only vectors). Not shipped to users and CI already scopes to prod deps —
+  but the fix is a routine **vite 5→6/7 upgrade**. Effort S.
+- **Duplication (L-5):** both `MailKit` (Infrastructure) and the frontend carry sizeable deps;
+  `react-grid-layout` exists solely for the draggable dashboard the redesign plans to retire —
+  candidate for removal with the Analytics migration. No abandoned packages spotted.
+- CI supply-chain posture: gitleaks (full history) + NuGet/npm audits are **required checks** —
+  good. No SAST/CodeQL (L-6, nice-to-have).
+
+## 4. Error handling & reliability
+- **Good:** global `UseExceptionHandler` + RFC7807 ProblemDetails (no stack traces to clients);
+  Polly retry/backoff on Gmail sync; idempotent sync upserts; AI calls wrapped in try/catch with
+  non-AI fallbacks ("sync must never fail because the AI provider is down").
+- **M-6 · Known EF model warning:** `Email` has a global query filter but is the required end
+  of the `Email↔EmailLabel` relationship — logged on every boot; can yield surprising results
+  when filtered parents are excluded. Fix: matching filter on `EmailLabel` (one line) + test.
+  (Long-noted in logs; this is the nudge to actually do it.)
+- **L-7 · Multi-step writes without explicit transactions:** e.g. `GoogleAuthEvents` upsert and
+  sync batches rely on EF's single-SaveChanges transactionality — mostly fine; the sync
+  cursor-advance + batch-upsert pairing is the one place an explicit transaction would guard a
+  mid-batch crash (currently self-heals via idempotent re-sync — acceptable).
+
+## 5. Performance
+- **Fixed this cycle (verified live):** relevance-ranked FTS with weighted tsvector + GIN;
+  trigram GIN indexes for the previously non-sargable sender/domain `.Contains()`; HNSW vector
+  index ready for semantic search.
+- **Remaining (all Low, roadmapped):** offset pagination degrades on deep pages (keyset planned
+  in `docs/discovery/05`); frontend ships one **678 KB JS bundle** (no code-splitting — vite
+  `manualChunks`/dynamic import, effort S); no HTTP caching/ETags on read-heavy endpoints
+  (sidebar counts are fetched per page-load); `docker-compose` frontend nginx lacks gzip/brotli
+  confirmation for the bundle.
+- No N+1 patterns found (queries project with joins; aggregates precomputed in
+  `AnalyticsAggregate`).
+
+## 6. Code quality & architecture
+- Clean Architecture discipline is genuinely observed (dependency rule intact; thin controllers;
+  DTO mapping at boundaries). Config is env-driven throughout.
+- **L-8:** `AiService.GenerateQueryAsync` returns raw LLM output as a search query (advisory-only,
+  becomes a search string — harmless today; keep it that way when NL search lands: model output
+  must stay data, never an executable/action).
+- Dead-ish code: none significant; `docs/specs/*` and discovery docs are current.
+
+## 7. Testing & CI
+- **41 tests** (unit: parser/guard/unsubscribe/embeddings-null; integration: authz challenge,
+  tenant isolation, pagination clamp, search fallback ordering). Critical security invariants
+  (IDOR filter, SSRF guard, confirmed-destructive) **are tested** — better than most.
+- **Gap (M-7):** no live-Postgres test harness — FTS/ranking/fuzzy/pgvector paths are verified
+  manually against staging (documented in commit messages) but not repeatably in CI. A
+  Testcontainers-Postgres (pgvector image) job would convert those throwaway verifications into
+  permanent regression tests. Effort M.
+- Gap (L-9): no coverage for `CleanupService`/`SyncService` beyond compilation; no E2E of the
+  OAuth flow (hard without creds — acceptable).
+- CI: build+test+secrets+deps as **required** PR checks; auto-deploy to staging with a
+  post-deploy health gate; prod is tag-gated. Solid. Missing: lint/format check in CI (the
+  pre-commit hook enforces locally; add `dotnet format --verify-no-changes` job, effort S) (L-10).
+
+---
+
+## Prioritized remediation plan
+
+| # | Finding | Sev | Fix effort | Blast radius |
+|---|---------|-----|-----------|--------------|
+| 1 | **H-1** Enable FluentValidation auto-validation + 400 tests | High | S | Isolated (1 line + tests; verify clients) |
+| 2 | **H-2** Rate limiting (global + auth/export/unsub/AI policies) | High | S–M | Isolated (Program.cs + tests) |
+| 3 | **H-3** Data-at-rest posture: document now; retention setting + purge job + account-deletion; volume-encryption guidance; (defer field-level encryption decision) | High | S (doc) → M (retention) → L (encryption) | Doc: none · Retention: moderate · Encryption: large/architectural |
+| 4 | **M-3** Protect Data Protection keys at rest | Med | S–M | Isolated + key migration |
+| 5 | **M-1** Absolute session lifetime (+ revocation groundwork) | Med | S | Isolated |
+| 6 | **M-4** Remove default DB password; fail fast | Med | S | Isolated |
+| 7 | **M-2** CSP + security headers on the SPA nginx | Med | S | Isolated (needs SPA smoke test) |
+| 8 | **M-6** Fix EF query-filter warning (EmailLabel filter) | Med | S | Isolated + test |
+| 9 | **M-5** Vite upgrade (dev-dep CVEs) | Med | S | Frontend build only |
+| 10 | **M-7** Testcontainers live-Postgres CI job | Med | M | CI + new test project wiring |
+| 11 | L-2/L-3/L-4/L-10 (AllowedHosts, SMTP log level, DB TLS note, CI format check) | Low | S each | Isolated |
+| 12 | L-5 bundle split / grid-layout removal (with redesign) | Low | S–M | Frontend |
+
+---
+
+# Phase 2 — Remediation status (2026-07-02, all items approved & implemented)
+
+Shipped as four PRs (#20–#23), each with tests, green CI, and a verified staging deploy.
+
+| Finding | Status | How it was resolved | PR |
+|---------|--------|---------------------|----|
+| **H-1** validators never ran | ✅ **Resolved** | `AddFluentValidationAutoValidation()`; invalid DTOs now 400 at the boundary. Proven by 2 integration tests (pageSize=0, From>To → 400) via a new test-auth scheme | #20 |
+| **H-2** no rate limiting | ✅ **Resolved** | Global 300/min per-user (per-IP anonymous) + `auth` 10/min + `expensive` 20/min (export/unsubscribe/AI), config-driven, 429/no-queue. Proven by a 429-on-3rd-request test | #20 |
+| **H-3** plaintext bodies + unbounded retention | ✅ **Resolved (as scoped)** | `SECURITY.md` documents the deliberate posture (volume-encryption + backup guidance, delete-my-data procedure); **opt-in retention** (`DataRetention:*`, default off) + daily purge worker with 3 tests. Field-level encryption deliberately deferred (FTS can't index encrypted columns) — revisit before any multi-user deployment | #22 |
+| **M-1** sliding-only sessions | ✅ **Resolved** | Absolute 30 d cap (`Auth:AbsoluteSessionDays`) via issued-at stamp checked in `OnValidatePrincipal`; 3 unit tests (incl. missing-stamp = expired) | #20 |
+| **M-2** no CSP on the SPA | ✅ **Resolved** | Full CSP (`script-src 'self'`, `frame-ancestors 'none'`, …) + nosniff/XFO/Referrer-Policy + gzip on the SPA nginx; inline theme script moved to `/theme-init.js` to keep `script-src 'self'` honest. **Verified serving on staging** | #21 |
+| **M-3** unprotected DP key ring | ✅ **Resolved (opt-in)** | `DataProtection:CertificatePath/Password` → `ProtectKeysWithCertificate`; documented in SECURITY.md as recommended for shared hosts | #22 |
+| **M-4** default DB password | ✅ **Resolved** | Guessable default removed from `appsettings.json`; startup fails fast with a clear message when the connection string has no password | #20 |
+| **M-5** dev-dep esbuild CVEs | ✅ **Resolved** | vite 5→8 + plugin-react 6; `npm audit` now clean **including dev deps**; build verified | #21 |
+| **M-6** EF query-filter warning | ✅ **Resolved** | Matching tenant filter on `EmailLabel` (via Email navigation); boot warning confirmed gone from staging logs; cross-user invisibility test added | #20 |
+| **M-7** no live-Postgres tests | ✅ **Resolved** | CI `db-tests` job with a `pgvector/pg16` service container runs 3 permanent `Category=LiveDb` regression tests (FTS weighting, ts_headline sentinels, trigram typo fallback, pgvector cosine). **Confirmed green on the actual runner** | #23 |
+| **L-2** AllowedHosts `*` | ✅ Documented | Production checklist item in SECURITY.md (set at deployment) | #22 |
+| **L-3** SMTP recipient at Info | ✅ **Resolved** | Downgraded to Debug | #20 |
+| **L-4** DB TLS note | ✅ Documented | SECURITY.md: add `SSL Mode=Require` if Postgres ever leaves the host | #22 |
+| **L-10** no CI format gate | ✅ **Resolved** | `format` CI job (`dotnet format --verify-no-changes`) — `--no-verify` pushes can no longer bypass formatting | #23 |
+| L-1 CSRF (accepted risk) | ✅ Documented | SECURITY.md rationale (SameSite=Lax + CORS + JSON) with revisit conditions | #22 |
+| L-5/L-6/L-7/L-8/L-9 | ⏸ Deferred by design | Bundle-split & grid-layout removal ride the UI redesign; SAST/CodeQL, sync transaction hardening, and broader service coverage are Phase 3 recommendation candidates | — |
+
+**Test suite: 40 → 54 tests** (48 always-on + 3 retention + 3 live-DB in CI).
+Every deploy through the pipeline stayed green; staging verified after each batch.

CHANGELOG.md

@@ -0,0 +1,36 @@
+# Changelog
+
+All notable changes to InboxIntel are documented here. Format follows
+[Keep a Changelog](https://keepachangelog.com/); versions follow
+[Semantic Versioning](https://semver.org/). See [docs/WORKFLOW.md](docs/WORKFLOW.md).
+
+## [Unreleased]
+### Security
+- Patched all High/Moderate NuGet advisories the new dependency gate surfaced:
+  `System.Text.Json` 8.0.0→8.0.6, `Microsoft.Extensions.Caching.Memory` 8.0.0→8.0.1
+  (+ `DependencyInjection.Abstractions`→8.0.2), `System.Security.Cryptography.Xml`
+  8.0.1→8.0.3 (transitive pins), and `MailKit`/`MimeKit` 4.13.0→4.17.0.
+
+### Added
+- CI/CD pipeline (`.gitea/workflows/`): `security` (gitleaks secret scan + NuGet/npm
+  vulnerability gate), `deploy-staging` (auto-redeploy local staging on `develop`,
+  with a post-deploy health gate), `deploy-prod` (tag-gated production promotion,
+  inactive until the server exists). Verified end-to-end on the self-hosted runner.
+- Formal Git workflow & environment strategy (`docs/WORKFLOW.md`).
+- Staging environment overlay (`docker-compose.staging.yml`) — production-shaped
+  Linux containers on Windows, isolated ports/volumes.
+- Version-controlled Git hooks (`scripts/git-hooks/`) + installer
+  (`scripts/install-hooks.ps1`): pre-commit secret/format checks, pre-push
+  build+test gate.
+- `VERSION` file as the single source of truth for the release number.
+
+## [0.1.0] — scaffold
+### Added
+- .NET 8 Clean Architecture backend (Domain/Application/Infrastructure/Api) + React/Vite SPA.
+- Docker Compose stack (Postgres 16, API, frontend, optional nginx proxy).
+- Gitea Actions CI (backend build+test, frontend build) on `main`/`develop` + PRs.
+- Security hardening: encrypted OAuth tokens, EF global query filters (IDOR),
+  loopback binds, non-root containers, SSRF egress guard.
+- One-command deploy scripts (`deploy/up.ps1`, `deploy/up.sh`).
+
+[Unreleased]: https://your-gitea-host/InboxIntel/compare/v0.1.0...HEAD

Directory.Build.props

@@ -1,6 +1,6 @@
 <Project>
   <PropertyGroup>
-    <TargetFramework>net8.0</TargetFramework>
+    <TargetFramework>net10.0</TargetFramework>
     <Nullable>enable</Nullable>
     <ImplicitUsings>enable</ImplicitUsings>
     <LangVersion>latest</LangVersion>

RECOMMENDATIONS.md

@@ -0,0 +1,121 @@
+# InboxIntel — Phase 3 Recommendations
+
+**Date:** 2026-07-02 · Follows the completed audit remediation ([AUDIT_REPORT.md](AUDIT_REPORT.md)).
+Each item: what · concrete benefit · effort (S/M/L) · risk · sources. **Ranked by value-to-effort.**
+
+> Research notes: grounded in official primary sources (fetched 2026-07-02) plus the
+> competitor/feature research already performed in `docs/discovery/02-competitor-analysis.md`.
+> (Live web *search* was quota-limited this session; the load-bearing facts below — support
+> dates, EF 10 features, Npgsql 10, Renovate/Gitea — were verified against primary docs.)
+
+---
+
+## 1. Migrate .NET 8 → .NET 10 LTS ⚠️ deadline-driven
+- **What:** move the solution to .NET 10 / EF Core 10 / Npgsql provider 10; bump
+  `Pgvector.EntityFrameworkCore` off the 0.2.0 EF8-pin at the same time.
+- **Why (hard fact):** **.NET 8 support ends 2026-11-10 — ~4 months away.** After that: no
+  security patches. .NET 10 is LTS until Nov 2028. This is not optional, only *when*.
+- **Bonus value:** EF 10 brings **named query filters** (exactly our multi-filter tenant
+  scenario — e.g. tenant + soft-delete filters, selectively ignorable), **redacted inlined
+  constants in SQL logs** (privacy win for an email app), first-class `LeftJoin`, and
+  better parameterized-collection SQL (plan-cache friendly).
+- **Watch-outs:** Npgsql 10 changes `array.Contains(x)` translation to `= ANY(...)` (check
+  our GIN-indexed paths); the migration touches every csproj + CI images + Dockerfiles.
+  The live-DB CI job we just added is the safety net for the search paths.
+- **Effort: M** (mechanical + verify) · **Risk: M** · Blast radius: whole repo, but staged
+  behind the pipeline.
+- Sources: [.NET support policy](https://dotnet.microsoft.com/en-us/platform/support/policy/dotnet-core),
+  [EF Core 10 what's-new](https://learn.microsoft.com/en-us/ef/core/what-is-new/ef-core-10.0/whatsnew),
+  [Npgsql EF 10 release notes](https://www.npgsql.org/efcore/release-notes/10.0.html).
+
+## 2. Automated dependency updates via Renovate (self-hosted Gitea)
+- **What:** run Renovate against the Gitea instance (PAT with repo/user/issue scopes,
+  `platform=gitea`); it opens update PRs that ride the existing required CI gates
+  (build, tests, gitleaks, vuln scan, live-DB).
+- **Benefit:** closes the audit's supply-chain gap permanently — the MailKit/System.Text.Json
+  CVE round we did by hand becomes an automated PR you just merge. NuGet + npm + Dockerfile
+  + Actions all covered.
+- **Effort: S** (a config + a scheduled runner job) · **Risk: L** (PRs are gated by CI).
+- Source: [Renovate Gitea platform docs](https://docs.renovatebot.com/modules/platform/gitea/).
+
+## 3. Database backups (currently none!)
+- **What:** nightly `pg_dump` sidecar/cron in compose, rotating N days, written to a
+  host path covered by your disk-encryption/backup regime (per SECURITY.md).
+- **Benefit:** today a bad migration or volume loss = total data loss; the audit fixed
+  security but the **availability** story is a single Docker volume. Highest
+  value-per-line-of-config item on this list.
+- **Effort: S** · **Risk: L**. Pair with a documented restore drill.
+
+## 4. Activate semantic search (Ollama + embedding backfill + hybrid ranking)
+- **What:** the pgvector column, HNSW index, `IEmbeddingProvider`, and live-DB tests are
+  already shipped. Remaining: an optional `ollama` compose profile, the embedding
+  backfill worker (batched, VRAM-aware), and RRF hybrid merge in `SearchService`
+  (design: `docs/discovery/05/06`).
+- **Benefit:** the flagship differentiator from the discovery blueprint — *"gym receipt
+  march"* finds the email; nobody mainstream offers this locally/privately.
+- **Effort: M–L** · **Risk: M** (quality tuning) · Needs the RTX-3080 box to pull
+  `nomic-embed-text` (~0.5 GB, always-on per the AI strategy).
+
+## 5. Observability: OpenTelemetry + a dashboard
+- **What:** wire .NET's built-in OTel (traces/metrics for ASP.NET, EF, HttpClient) exported
+  to a compose-profile Prometheus+Grafana (or an OTLP endpoint later). Keep Serilog for logs.
+- **Benefit:** today diagnosis = `docker logs`. This gives request latency, sync-job
+  timings, rate-limit hits, and search-performance baselines — the "what will break first
+  as usage grows" early-warning system.
+- **Effort: M** · **Risk: L** (additive).
+
+## 6. Named query filters for tenancy (after #1)
+- **What:** convert the hand-rolled `CurrentUserId == Guid.Empty || …` filters to EF 10
+  named filters (`"Tenant"`, future `"SoftDelete"`), selectively ignorable per-query.
+- **Benefit:** cleaner + safer than the worker-bypass convention; directly feeds the
+  multi-provider platform's isolation model.
+- **Effort: S** (post-migration) · **Risk: L** (isolation tests already exist).
+
+## 7. Frontend bundle code-splitting
+- **What:** vite `manualChunks`/dynamic imports to split the 678 KB bundle (charts,
+  grid-layout, per-route chunks); drop `react-grid-layout` when the Analytics redesign
+  lands (it's the sole consumer).
+- **Benefit:** faster cold loads; audit L-5 closed. Gzip is already on (batch B), so this
+  is the remaining lever.
+- **Effort: S–M** · **Risk: L**.
+
+## 8. Keyset (cursor) pagination for search
+- **What:** replace offset `Skip/Take` with keyset pagination for browse/date-ordered
+  paths; ranked paths already effectively top-N (design in `docs/discovery/05`).
+- **Benefit:** deep-page latency stops degrading linearly at 100k+ mailboxes.
+- **Effort: M** (API shape + frontend infinite-scroll cursor) · **Risk: M** (API change).
+
+## 9. SAST in CI (Semgrep)
+- **What:** a `semgrep` job (OSS rules for C#/JS + secrets/OWASP packs) in `ci.yml` —
+  CodeQL is GitHub-centric; Semgrep runs anywhere Docker does.
+- **Benefit:** closes audit L-6; catches injection/crypto misuse patterns the current
+  gates (gitleaks + dep-audit + tests) don't look for.
+- **Effort: S–M** (tuning noise) · **Risk: L** (advisory job first, required later).
+
+## 10. Settings + feature-flag platform (multi-provider Phase 4)
+- **What:** implement `docs/discovery/multi-provider/04` — `feature_flags`/`user_settings`
+  tables, `IFeatureFlags`/`IAiGate`, admin toggles later.
+- **Benefit:** unblocks shipping AI features dark (`ai.enabled` master switch), the
+  Settings UI, and everything in the multi-provider plan; prerequisite for #4 to be
+  properly gated per the approved design.
+- **Effort: L** · **Risk: M** — the biggest item, but the one the roadmap already commits to.
+
+---
+
+## Value-to-effort ranking (summary)
+| # | Item | Effort | Why this rank |
+|---|------|--------|---------------|
+| 1 | .NET 10 migration | M | **EOL deadline Nov 2026**; unlocks #6 |
+| 2 | Renovate | S | Permanent supply-chain automation for one config file |
+| 3 | Backups | S | Only protection against total data loss |
+| 4 | Semantic search activation | M–L | Flagship product differentiator; infra already live |
+| 5 | OpenTelemetry | M | Can't manage what you can't see |
+| 6 | Named query filters | S | Cheap once #1 lands |
+| 7 | Bundle splitting | S–M | Perceived speed; last audit-perf leftover |
+| 8 | Keyset pagination | M | Scales search; roadmap item |
+| 9 | Semgrep SAST | S–M | Last unautomated security layer |
+| 10 | Settings/flags platform | L | Roadmap-committed foundation |
+
+**Suggested sequencing:** 2+3 immediately (tiny, standalone) → 1 (deadline) → 6 → 4 (+10 gating if you want flags first) → 5 → 7/8/9 opportunistically.
+
+**STOP — awaiting your selections before implementing anything (Phase 4).**

SECURITY.md

@@ -0,0 +1,65 @@
+# InboxIntel — Security & Data Posture
+
+The deliberate security posture of this application, so operators know exactly what is and
+isn't protected. Complements [AUDIT_REPORT.md](AUDIT_REPORT.md) (point-in-time audit) and
+`docs/discovery/multi-provider/06-security-model.md` (future multi-user design).
+
+## Deployment model this posture assumes
+Self-hosted, **single-operator** instance: the person running the server is the person whose
+mailbox is synced. All services (API, Postgres, frontend) run in Docker on the operator's own
+machine; Postgres and the API bind to loopback/compose-internal only; TLS terminates at the
+reverse proxy.
+
+## What is protected, and how
+| Asset | Protection |
+|---|---|
+| Google OAuth refresh/access tokens | Encrypted at rest (ASP.NET Data Protection, AES); never logged; never sent to the browser |
+| Data Protection key ring | Optionally encrypted with an operator-supplied X.509 certificate — set `DataProtection:CertificatePath`/`CertificatePassword`. **Without it, keys sit in plaintext on the `/keys` volume** and anyone with volume access can decrypt stored tokens. Recommended for any shared host. |
+| App session | HttpOnly/SameSite=Lax/Secure cookie · sliding 7 d **with an absolute 30 d cap** (`Auth:AbsoluteSessionDays`) |
+| Login/abuse | Rate limiting: global 300 req/min per user (or IP when anonymous); `auth` 10/min; export/unsubscribe/AI 20/min (`RateLimiting:*`) |
+| Cross-user access | EF global query filters on every tenant-scoped entity (incl. the EmailLabel join) — tested |
+| Outbound fetches (unsubscribe etc.) | `SafeHttpGuard` SSRF allowlisting (DNS-rebinding-safe) + redirects disabled |
+| Untrusted email content in the UI | Rendered only as escaped React text (no `dangerouslySetInnerHTML`); search highlights use non-HTML sentinels; SPA ships CSP with `script-src 'self'` |
+| Secrets | Never committed (`deploy/.env*` git-ignored; pre-commit + CI gitleaks scans); no passwords stored at all (OAuth-only login) |
+
+## What is deliberately NOT protected (accepted risks — read this)
+1. **Email bodies are stored in plaintext in Postgres.** Full-text and semantic search index
+   the body; encrypted columns cannot be indexed this way. On the assumed single-operator
+   deployment, the database lives on the operator's own disk, so the threat this would
+   mitigate (a third party reading the DB files) reduces to "someone with access to your
+   machine" — mitigate it at the layer that actually works:
+   - **Use full-disk or volume encryption** on the host (BitLocker/LUKS) — strongly recommended.
+   - **Encrypt backups**: nightly `pg_dump` rotation runs via the compose `backup` service
+     into `./backups/` (git-ignored) — keep that directory on an encrypted disk and copy it
+     off-machine. Restore: `docker compose exec -T postgres psql -U inboxintel -d inboxintel < backups/<file>.sql`.
+   - Before any **multi-user** deployment, revisit per the multi-provider security design
+     (host admins must not be able to read members' mail — plaintext bodies break that promise).
+2. **DB connection is not TLS** — Postgres is only reachable on the compose-internal network /
+   loopback. If you ever move Postgres to another host, add `SSL Mode=Require` to the
+   connection string (audit L-4).
+3. **No CSRF tokens** — SameSite=Lax cookies + strict CORS + JSON-only bodies make classic
+   CSRF impractical; revisit if either changes (audit L-1).
+
+## Data retention (opt-in)
+By default the local mailbox copy is kept indefinitely. Two knobs enable automatic purging of
+the **local copy only** (your actual Gmail is never touched):
+```json
+"DataRetention": {
+  "PurgeTrashedAfterDays": 0,   // e.g. 30 — purge local copies of trashed mail after 30 days
+  "PurgeAllAfterDays": 0        // e.g. 730 — keep at most ~2 years of mail locally
+}
+```
+`0` disables a knob. A daily background worker applies them. For a full "delete my data"
+operation: stop the stack and remove the `pgdata` + `keys` volumes
+(`docker compose down -v`), and revoke the app's access in your Google account.
+
+## Production checklist (beyond compose defaults)
+- Set `AllowedHosts` to your real hostname(s) (audit L-2).
+- Terminate TLS at the proxy; HSTS is enabled automatically outside Development.
+- Provide `DataProtection:CertificatePath` to encrypt the key ring.
+- Keep `deploy/.env` readable only by the service user; rotate the DB password if exposed.
+- Dependency + secret scanning run in CI on every PR (required checks).
+
+## Reporting
+Single-operator project — if you find a vulnerability, open a private issue or contact the
+repository owner directly.

VERSION

@@ -0,0 +1 @@
+0.1.0

deploy/down.ps1

@@ -5,13 +5,22 @@
   ./deploy/down.ps1
   ./deploy/down.ps1 -Volumes
 #>
-param([switch]$Volumes)
+param([switch]$Volumes, [switch]$Staging)
 
 $ErrorActionPreference = 'Stop'
-$root    = Split-Path -Parent $PSScriptRoot
+$root = Split-Path -Parent $PSScriptRoot
-$envFile = Join-Path $PSScriptRoot '.env'
 
-$composeArgs = @('compose', '--env-file', $envFile, 'down')
+if ($Staging) {
+  $envFile      = Join-Path $PSScriptRoot '.env.staging'
+  $composeFiles = @('-f', 'docker-compose.yml', '-f', 'docker-compose.staging.yml')
+  $project      = @('-p', 'inboxintel-staging')
+} else {
+  $envFile      = Join-Path $PSScriptRoot '.env'
+  $composeFiles = @()
+  $project      = @()
+}
+
+$composeArgs = @('compose') + $project + @('--env-file', $envFile) + $composeFiles + @('down')
 if ($Volumes) { $composeArgs += '--volumes' }
 
 Push-Location $root

deploy/up.ps1

@@ -8,15 +8,28 @@
 #>
 param(
   [switch]$Proxy,
-  [switch]$Foreground
+  [switch]$Foreground,
+  [switch]$Staging   # production-shaped staging stack: separate env, ports, volumes
 )
 
 $ErrorActionPreference = 'Stop'
-$root    = Split-Path -Parent $PSScriptRoot   # repo root (deploy/ is one level down)
+$root = Split-Path -Parent $PSScriptRoot   # repo root (deploy/ is one level down)
-$envFile = Join-Path $PSScriptRoot '.env'
 
-if (-not (Test-Path $envFile)) {
+# Staging vs production: pick the env file + compose overlay + isolated project name.
-  throw "Missing $envFile. Create it from .env.example with your real secrets."
+if ($Staging) {
+  $envFile     = Join-Path $PSScriptRoot '.env.staging'
+  $composeFiles = @('-f', 'docker-compose.yml', '-f', 'docker-compose.staging.yml')
+  $project     = @('-p', 'inboxintel-staging')
+  if (-not (Test-Path $envFile)) {
+    throw "Missing $envFile. Create it from .env.staging.example."
+  }
+} else {
+  $envFile     = Join-Path $PSScriptRoot '.env'
+  $composeFiles = @()
+  $project     = @()
+  if (-not (Test-Path $envFile)) {
+    throw "Missing $envFile. Create it from .env.example with your real secrets."
+  }
 }
 
 # Fail fast if a required key is absent or blank.
@@ -28,19 +41,23 @@ Get-Content $envFile | ForEach-Object {
 $missing = $required | Where-Object { [string]::IsNullOrWhiteSpace($envMap[$_]) }
 if ($missing) { throw "deploy/.env is missing values for: $($missing -join ', ')" }
 
-$composeArgs = @('compose', '--env-file', $envFile)
+$composeArgs = @('compose') + $project + @('--env-file', $envFile) + $composeFiles
 if ($Proxy) { $composeArgs += @('--profile', 'proxy') }
 $composeArgs += @('up', '--build')
 if (-not $Foreground) { $composeArgs += '-d' }
 
 Push-Location $root
 try {
-  Write-Host "Starting InboxIntel via docker compose (env: deploy/.env)..." -ForegroundColor Cyan
+  Write-Host "Starting InboxIntel via docker compose (env: $envFile)..." -ForegroundColor Cyan
   & docker @composeArgs
   if (-not $Foreground) {
-    & docker compose --env-file $envFile ps
+    & docker compose @project --env-file $envFile @composeFiles ps
-    Write-Host "`nFrontend: http://localhost:8081   API/Swagger: http://localhost:8080/swagger" -ForegroundColor Green
+    if ($Staging) {
-    Write-Host "Logs: ./deploy/logs.ps1   Stop: ./deploy/down.ps1" -ForegroundColor DarkGray
+      Write-Host "`n[STAGING] Frontend: http://localhost:18081   API/Swagger: http://localhost:18080/swagger" -ForegroundColor Green
+    } else {
+      Write-Host "`nFrontend: http://localhost:8081   API/Swagger: http://localhost:8080/swagger" -ForegroundColor Green
+    }
+    Write-Host "Logs: ./deploy/logs.ps1   Stop: ./deploy/down.ps1$(if($Staging){' -Staging'})" -ForegroundColor DarkGray
   }
 }
 finally { Pop-Location }

docker-compose.staging.yml

@@ -0,0 +1,46 @@
+# Staging overlay for InboxIntel.
+#
+# The base docker-compose.yml IS the production definition (Linux containers,
+# ASPNETCORE_ENVIRONMENT=Production). This overlay layers a *staging* variant on
+# top of it so you can run a production-shaped stack locally on Windows WITHOUT
+# clobbering a real production deployment's data, ports, or volumes.
+#
+# It differs from prod only in the ways staging is meant to differ:
+#   - the dev/test banner is on (App__DevMode=true)
+#   - the initial Gmail sync is capped so a big mailbox doesn't take forever
+#   - ports are shifted into the 18xxx range so staging can run alongside prod
+#   - a distinct project name gives it its own isolated pgdata + keys volumes
+#
+# Run it with the -p (project name) flag so volumes/networks are namespaced:
+#
+#   docker compose -p inboxintel-staging \
+#     --env-file deploy/.env.staging \
+#     -f docker-compose.yml -f docker-compose.staging.yml up --build -d
+#
+# (deploy/up.ps1 -Staging / up.sh --staging wrap this for you.)
+
+services:
+  postgres:
+    # !override replaces the base port list instead of merging with it, so staging
+    # binds ONLY its shifted 18xxx/15432 ports and never squats on prod's 8080/8081.
+    ports: !override
+      - "127.0.0.1:15432:5432"
+
+  api:
+    environment:
+      # Same Production runtime as prod (real config binding, real build), but
+      # flagged as a non-production instance so the UI shows the staging banner
+      # and the first sync is bounded.
+      App__DevMode: "true"
+      GmailSync__MaxMessages: ${MAX_MESSAGES:-2000}
+      Cors__Origins__0: ${FRONTEND_ORIGIN:-http://localhost:18081}
+    ports: !override
+      - "127.0.0.1:18080:8080"
+
+  frontend:
+    ports: !override
+      - "18081:80"
+
+  nginx:
+    ports: !override
+      - "18000:80"

docker-compose.yml

@@ -1,6 +1,8 @@
 services:
   postgres:
-    image: postgres:16-alpine
+    # pgvector-enabled Postgres 16 (semantic search). Drop-in for postgres:16 data;
+    # the 'vector' extension is created by the AddEmbeddingColumn migration.
+    image: pgvector/pgvector:pg16
     environment:
       POSTGRES_DB: inboxintel
       POSTGRES_USER: inboxintel
@@ -20,6 +22,47 @@ services:
       timeout: 5s
       retries: 10
 
+  # Nightly logical backups (RECOMMENDATIONS #3 — previously there were NONE). Dumps
+  # rotate after BACKUP_KEEP_DAYS. The ./backups host directory should live on an
+  # encrypted disk and be included in your off-machine backup regime (see SECURITY.md).
+  # Restore: docker compose exec -T postgres psql -U inboxintel -d inboxintel < backups/<file>.sql
+  backup:
+    image: pgvector/pgvector:pg16
+    entrypoint: /bin/sh
+    command:
+      - -c
+      - |
+        while true; do
+          ts=$$(date -u +%Y%m%d-%H%M%S)
+          if pg_dump -h postgres -U inboxintel -d inboxintel > /backups/inboxintel-$$ts.sql.tmp; then
+            mv /backups/inboxintel-$$ts.sql.tmp /backups/inboxintel-$$ts.sql
+            echo "backup OK: inboxintel-$$ts.sql"
+          else
+            rm -f /backups/inboxintel-$$ts.sql.tmp
+            echo "backup FAILED at $$ts" >&2
+          fi
+          find /backups -name 'inboxintel-*.sql' -mtime +$${BACKUP_KEEP_DAYS:-7} -delete
+          sleep 86400
+        done
+    environment:
+      PGPASSWORD: ${POSTGRES_PASSWORD:?set POSTGRES_PASSWORD in deploy/.env}
+      BACKUP_KEEP_DAYS: ${BACKUP_KEEP_DAYS:-7}
+    volumes:
+      - ./backups:/backups
+    depends_on:
+      postgres:
+        condition: service_healthy
+
+  # One-shot: ensure the DataProtection 'keys' volume is owned by the API's non-root
+  # 'app' user (uid 1654). A volume created by an older root-running image is root-owned,
+  # which makes the app fail to read its key ring and 500s on login. Runs as root, chowns,
+  # exits; the api waits for it. Idempotent and cheap.
+  init-keys:
+    image: busybox
+    command: ["sh", "-c", "chown -R 1654:1654 /keys"]
+    volumes:
+      - keys:/keys
+
   api:
     build:
       context: .
@@ -32,6 +75,10 @@ services:
       GoogleOAuth__ClientId: ${GOOGLE_CLIENT_ID:-}
       GoogleOAuth__ClientSecret: ${GOOGLE_CLIENT_SECRET:-}
       Ai__Mode: ${AI_MODE:-Disabled}
+      # Points at the compose 'ollama' service when the ai profile is up; harmless otherwise.
+      Ai__OllamaBaseUrl: ${OLLAMA_BASE_URL:-http://ollama:11434}
+      # OTLP export activates only when set (e.g. http://lgtm:4317 with the observability profile).
+      OTEL_EXPORTER_OTLP_ENDPOINT: ${OTEL_ENDPOINT:-}
       # Dev mode shows the dev banner and caps the initial sync. Set DEV_MODE=true
       # and MAX_MESSAGES=1000 in deploy/.env to exercise it in this Docker setup.
       App__DevMode: ${DEV_MODE:-false}
@@ -42,6 +89,8 @@ services:
     depends_on:
       postgres:
         condition: service_healthy
+      init-keys:
+        condition: service_completed_successfully
     # V-08: bind to loopback so the API is not directly reachable from the network
     # (only via the frontend/nginx proxy over the internal compose network). This
     # prevents external clients from bypassing the proxy to spoof X-Forwarded-* headers.
@@ -57,6 +106,33 @@ services:
     ports:
       - "8081:80"
 
+  # Local AI (semantic search + assistants). Enable with:
+  #   docker compose --profile ai up -d && set AI_MODE=LocalOllama in deploy/.env
+  # First run: docker compose exec ollama ollama pull nomic-embed-text
+  # GPU (RTX 3080): uncomment the deploy block to pass the GPU through.
+  ollama:
+    image: ollama/ollama
+    profiles: ["ai"]
+    volumes:
+      - ollama:/root/.ollama
+    # deploy:
+    #   resources:
+    #     reservations:
+    #       devices:
+    #         - driver: nvidia
+    #           count: all
+    #           capabilities: [gpu]
+
+  # Observability (RECOMMENDATIONS #5): all-in-one Grafana+Tempo+Prometheus+Loki.
+  # Enable with: docker compose --profile observability up -d
+  # then set OTEL_ENDPOINT=http://lgtm:4317 in deploy/.env and restart the api.
+  # Grafana UI: http://localhost:3000 (admin/admin on first run).
+  lgtm:
+    image: grafana/otel-lgtm
+    profiles: ["observability"]
+    ports:
+      - "127.0.0.1:3000:3000"
+
   # Optional reverse proxy. Enable with: docker compose --profile proxy up
   nginx:
     image: nginx:alpine
@@ -72,3 +148,4 @@ services:
 volumes:
   pgdata:
   keys:
+  ollama:

docs/WORKFLOW.md

@@ -0,0 +1,188 @@
+# InboxIntel — Git Workflow & Environment Strategy
+
+The single source of truth for how we branch, commit, version, and promote code
+across environments. Optimised for a **solo developer on Windows 11 with a future
+Linux production server**. Kept deliberately lightweight — every rule here earns
+its place.
+
+---
+
+## 1. Branch strategy
+
+A trimmed **GitHub Flow + a long-lived `develop`** model. Two permanent branches,
+short-lived branches off `develop`.
+
+| Branch          | Lives for | Purpose                                              | Deploys to |
+|-----------------|-----------|------------------------------------------------------|------------|
+| `main`          | forever   | Always releasable. Every commit is tagged & shippable | production |
+| `develop`       | forever   | Integration branch. What staging runs                 | staging    |
+| `feature/*`     | hours–days| One feature or refactor                               | dev (local)|
+| `fix/*`         | hours     | Non-urgent bug fix                                    | dev (local)|
+| `hotfix/*`      | minutes–hrs| Urgent prod fix, branched from `main`                | prod (fast)|
+| `release/x.y.0` | optional  | Only if a release needs stabilisation before tagging  | staging    |
+
+**Why this shape (not full GitFlow):** a solo dev doesn't need GitFlow's ceremony
+(separate release managers, parallel release trains). But keeping `develop`
+separate from `main` gives one thing that pure trunk-based can't: a **staging
+environment that always mirrors `develop`** while `main` stays clean and
+tag-perfect for production. `release/*` exists only when you want to freeze
+features and stabilise — skip it for routine work.
+
+### Normal flow
+
+```
+main ──────●────────────────────●──────────  (tagged: v1.2.0, v1.3.0)
+            \                   /
+develop ──●──●──●──●──●──●──●──●────────────  (staging)
+              \   /   \       /
+feature/x      ●─●     \     /
+fix/y                   ●───●
+```
+
+1. `git switch develop && git pull`
+2. `git switch -c feature/sender-policy`
+3. Commit in small conventional commits.
+4. Push, open a PR **into `develop`** (Gitea). CI must be green.
+5. Squash-merge. Delete the branch.
+6. When `develop` is ready to ship → PR `develop → main`, tag, deploy.
+
+---
+
+## 2. Commit conventions — Conventional Commits
+
+Format: `type(scope): short imperative summary`
+
+**Types:** `feat`, `fix`, `docs`, `refactor`, `test`, `chore`, `ci`, `perf`,
+`build`, `style`, `security`.
+
+Rules:
+- Summary ≤ 72 chars, imperative mood ("add", not "added").
+- One logical change per commit.
+- Body explains **why**, not what (the diff shows what).
+- Breaking change → add `!` (`feat!:`) and a `BREAKING CHANGE:` footer.
+
+Examples (matching this repo's history):
+```
+feat(ui): F1+F2 — component primitives + rebuilt app shell
+fix(security): systemic IDOR safeguard via EF global query filters
+ci: add Gitea Actions build + test pipeline
+```
+
+Why: conventional commits drive **automatic semver bumps** and a generated
+CHANGELOG, and make `git log` scannable. `feat` → minor, `fix` → patch,
+`BREAKING CHANGE` → major.
+
+---
+
+## 3. Pull request / merge flow (Gitea)
+
+Even solo, PRs are worth it: they run CI, give a diff review checkpoint, and build
+a paper trail.
+
+- **Target:** `feature/*` and `fix/*` → `develop`. `develop`/`hotfix/*` → `main`.
+- **Gate:** the `CI` workflow (backend build+test, frontend build) must pass.
+- **Merge style:** **squash-merge** feature branches (one clean commit on
+  `develop`). **Merge commit** for `develop → main` (preserves the integration
+  history and makes the release boundary visible).
+- **Branch protection (Gitea → Settings → Branches):** protect `main` and
+  `develop`; require status checks to pass; disallow force-push.
+
+---
+
+## 4. Versioning — Semantic Versioning (`MAJOR.MINOR.PATCH`)
+
+- **MAJOR** — breaking API/behaviour change.
+- **MINOR** — backward-compatible feature.
+- **PATCH** — backward-compatible fix.
+
+The current version lives in [`VERSION`](../VERSION) and is the one place bumped
+per release. Pre-1.0 while scaffolding: stay on `0.x` (minor = features, patch =
+fixes; anything may change).
+
+---
+
+## 5. Tagging & releases
+
+Tags are cut **only on `main`**, annotated, prefixed `v`:
+
+```bash
+git switch main && git pull
+# bump VERSION + CHANGELOG in a release commit, then:
+git tag -a v1.3.0 -m "v1.3.0 — sender policy + staging overlay"
+git push origin main --tags
+```
+
+- Tag == the exact commit deployed to production == the rollback target.
+- The tag message summarises the release; details live in `CHANGELOG.md`.
+
+---
+
+## 6. Hotfix process
+
+For a bug already in production:
+
+```bash
+git switch main && git pull
+git switch -c hotfix/oauth-callback-500
+# fix + test
+git commit -m "fix(security): guard null OAuth state on callback"
+# PR hotfix/* -> main, CI green, merge
+git switch main && git pull
+git tag -a v1.3.1 -m "v1.3.1 hotfix — OAuth callback" && git push origin main --tags
+# deploy the tag, THEN back-merge so develop doesn't lose the fix:
+git switch develop && git merge main && git push
+```
+
+The **back-merge to `develop`** is the step people forget — without it the next
+release silently reintroduces the bug.
+
+---
+
+## 7. Environments
+
+| Env         | Host                    | Runtime                              | Config source            | Purpose            |
+|-------------|-------------------------|--------------------------------------|--------------------------|--------------------|
+| Development | Windows 11 (native)     | `dotnet run` + `vite` (hot reload)   | `appsettings.Development.json` + user-secrets | fast iteration, debugging |
+| Staging     | Windows 11 (Docker)     | Linux containers, `ASPNETCORE_ENVIRONMENT=Production` | `deploy/.env.staging` + `docker-compose.staging.yml` | production rehearsal |
+| Production  | Linux server (Docker)   | identical Linux containers            | `deploy/.env` on the server | live |
+
+**Parity principle:** dev is fast (native, Windows) and *not* production-shaped —
+that's fine, it's for inner-loop speed. **Staging is the parity gate**: it runs the
+*same Linux images* Docker builds for production, so "works in staging" genuinely
+predicts "works in prod". The only staging↔prod differences are the dev banner,
+capped sync, ports, and volume namespace — see `docker-compose.staging.yml`.
+
+### Commands
+
+```powershell
+# DEV (native, hot reload) — two terminals
+dotnet run --project src/InboxIntel.Api        # http://localhost:5080
+cd frontend; npm run dev                        # http://localhost:5173
+
+# STAGING (production-shaped, on Windows via Docker)
+./deploy/up.ps1 -Staging                         # http://localhost:18081
+./deploy/down.ps1 -Staging
+
+# PRODUCTION-shaped run locally (smoke test the real config)
+./deploy/up.ps1                                  # http://localhost:8081
+```
+
+Dev and staging use **different ports and different volumes**, so they can run at
+the same time and never share a database.
+
+---
+
+## 8. Git hooks
+
+Installed via `core.hooksPath` (run once per clone):
+
+```powershell
+./scripts/install-hooks.ps1
+```
+
+- **pre-commit** (fast): blocks committed `.env`/secrets, verifies `dotnet format`.
+- **pre-push** (thorough): `dotnet build -c Release` + `dotnet test` + frontend
+  build — the same gate CI runs, caught before the push.
+
+Emergency bypass: `--no-verify`. CI still enforces the gate server-side, so a
+bypassed push can still be rejected by branch protection.

docs/discovery/00-design-brief.md

@@ -0,0 +1,45 @@
+# Design Brief (locked via interview)
+
+The confirmed design direction from the Phase 4A interview. This governs the UX/UI
+redesign, design system, themes, and search experience.
+
+## Direction in one line
+**An approachable, professional, "anyone can pick it up" email tool in the spirit of
+Notion & Arc — a polished, responsive web app that is fast and premium without being
+intimidating.**
+
+## Confirmed preferences
+| Dimension | Decision | Design implication |
+|-----------|----------|--------------------|
+| Reference feel | **Notion / Arc, but professional** | Friendly, spatial, low learning curve — not a power-user speed tool (not Superhuman). |
+| Density | **Balanced** | Comfortable, readable rows; efficient but not cramped. |
+| Input model | **Pointer-first / discoverable** | Clickable UI is the backbone. Command palette + shortcuts are **accelerators for power users**, layered on top — never required. |
+| App metaphor | **Polished web app** | Exceptional in-browser experience; shareable URLs; no native-desktop assumptions. |
+| Platforms | **Desktop (Win + Mac/Linux), Tablet, Mobile** | **Fully responsive** is mandatory — layouts must gracefully collapse desktop → mobile. |
+| Default theme | **Dark-first** | Design primarily for a layered dark UI; light theme fully first-class (genuine white, not grey). |
+| Motion | **Subtle & smooth** | Gentle, quick transitions and tasteful micro-interactions. Respect `prefers-reduced-motion`. |
+| Colour personality | **Warm ∩ cool blend, green accent** | Accent built around **`#3ba31f`** (refined into an accessible ramp); neutrals lean subtly warm. |
+| Icons | **Clean line, rounded** (Lucide-style) | Consistent stroke weight, soft corners. |
+| Speed vs polish | **Balanced** | Polish is welcome only when it never costs perceptible speed. |
+| Audience | **Mainstream, power-capable** | Simple by default; power features revealed progressively. |
+| Accessibility | **Deferred (nice-to-have)** | AA basics baked in cheaply; deeper a11y is a pre-launch backlog item. See below. |
+
+## The pivotal insight
+The original brief leaned "keyboard-first"; the interview corrected this to
+**discoverable-first**. That reframes the flagship **search** away from a syntax you
+must learn (`from:x after:y`) toward an **inviting, visual, assisted** experience —
+filter chips, live suggestions, and natural language — with the power syntax still
+available underneath. This single decision shapes the entire redesign.
+
+## Accent colour note
+`#3ba31f` = rgb(59,163,31). It's a strong differentiator (most email apps default to
+blue) and bridges warm/cool. It will be developed into a full ramp (50→900); the
+**interactive** shade will be nudged for WCAG AA contrast on dark surfaces and on
+buttons, and green will **never be the sole signal** for selection/status (always
+paired with icon/shape) so the design is colour-blind-safe by construction.
+
+## Deferred: accessibility
+Per the user, accessibility is a **nice-to-have to revisit before public launch**, not
+a hard requirement now. Cheap AA basics (contrast, focus rings, reduced-motion,
+non-colour-only state) are still baked in; high-contrast mode, full screen-reader
+semantics, font-scaling controls, and formal colour-blind audits are **backlog**.

docs/discovery/01-architecture-review.md

@@ -0,0 +1,88 @@
+# 01 — Architecture Review (current state)
+
+Phase 1 deliverable: a grounded assessment of InboxIntel as it exists today, from the
+source. This is the baseline the redesign builds on.
+
+## Architecture
+**Clean Architecture, .NET 8** — strict dependency rule `Api → Infrastructure → Application → Domain`.
+
+| Layer | Responsibility | Key contents |
+|-------|----------------|--------------|
+| **Domain** | Entities + enums, no external deps | `Email`, `Sender`, `Domain`, `MailThread`, `Label`/`EmailLabel`, `Attachment`; `EmailCategory` |
+| **Application** | Interfaces, DTOs, validation, parsing | `ISearchService` et al., `GmailQueryParser`, FluentValidation |
+| **Infrastructure** | EF Core/Npgsql, integrations, workers | `SearchService`, `CleanupService`, `AnalyticsService`, `HeuristicClassifier`, **AI providers**, exports, `GmailSyncWorker`, `DigestWorker` |
+| **Api** | ASP.NET Core Web API | Thin controllers, DI, Serilog, OAuth |
+| **Frontend** | React 18 + Vite SPA | Chart.js, react-grid-layout (draggable dashboard), Tailwind |
+
+- **Data:** PostgreSQL (EF Core + Npgsql). `Email` designed for 100k+ rows/user; a
+  **generated `tsvector`** column backs full-text search.
+- **Background:** hosted workers — `GmailSyncWorker` (scheduled sync), `DigestWorker` (digest).
+- **AI seam (already present):** `IAiProvider` with `NullAiProvider` (disabled),
+  `OllamaProvider` (local `/api/chat`), `OpenAiProvider` (cloud). Contract today is a
+  single `CompleteAsync(systemPrompt, userPrompt)`.
+- **Security:** Google OAuth2 (read-only Gmail), cookie session + JWT, refresh tokens
+  encrypted via Data Protection API, **IDOR-safe global query filters**, SSRF egress
+  guard, non-root containers, destructive actions require `Confirmed` + server preview.
+- **Delivery:** Docker Compose (Postgres/API/frontend/optional nginx) + Gitea CI/CD.
+
+## Feature set
+Gmail connect + sync → Postgres · analytics dashboard (draggable widgets) · advanced
+search (Gmail operators + FTS) · safe bulk cleanup · unsubscribe management
+(List-Unsubscribe / one-click) · heuristic categorisation (smart folders) · exports
+(PDF/CSV/JSON) · optional **advisory** AI · sender/domain aggregation.
+
+## Existing search implementation (flagship)
+1. **`GmailQueryParser`** (Application) parses `from: to: domain: after: before:
+   is:unread|read has:attachment`; remaining text → free-text.
+2. **`SearchService`** (Infrastructure) composes structured filters as EF `WHERE`
+   clauses, and free text via Postgres FTS:
+   `SearchVector.Matches(PlainToTsQuery('english', term))`, where
+   `SearchVector = to_tsvector('english', coalesce(Subject,'') || ' ' || coalesce(BodyText,''))`.
+3. Results are per-user filtered (IDOR-safe), **ordered by `SentAtUtc DESC`**,
+   offset-paginated, projected to `EmailSummaryDto`.
+
+## Strengths
+- Clean, testable layering; DI throughout; 39 tests + CI/CD gate.
+- **Real Postgres FTS** (generated tsvector), not naïve `LIKE`.
+- Security-forward (IDOR filters, encrypted tokens, SSRF guard, confirmed destructive ops).
+- **AI already decoupled behind `IAiProvider`** — the "no-AI / Ollama / future" requirement is architecturally seeded.
+- Read-only scope + advisory AI = privacy-respecting.
+
+## Weaknesses
+- **Search is single-mode**: ordered by **date, not relevance** (no `ts_rank`); no
+  fuzzy/typo tolerance, no semantic/vector search, no grouping, no "why matched," no
+  saved/recent/suggested searches.
+- FTS is **English-only** and covers **only Subject + Body** (not sender, attachment
+  names, labels).
+- Sender/domain filters use `.Contains()` → **non-sargable ILIKE scans** (no `pg_trgm`).
+- Category is a **single heuristic enum** — no multi-label, confidence, or learning.
+- **`IAiProvider` is chat-only** — no embeddings/classification/extraction contract, so
+  semantic search & structured extraction can't yet be expressed.
+- Threading (`ThreadId`) exists but isn't surfaced as conversation intelligence.
+
+## Technical debt
+- EF global-query-filter vs required `Email↔EmailLabel` relationship warning (in logs).
+- Hardcoded English FTS config.
+- **678 KB single JS chunk** (no code-splitting) — cold-load cost.
+- Relevance-blind ordering.
+- AI interface too narrow for the roadmap.
+
+## Performance bottlenecks
+- `.Contains()` sender/domain → sequential scans (need `pg_trgm` GIN indexes).
+- **Offset pagination** (`Skip/Take` + `Count`) degrades on deep pages → keyset/cursor.
+- No score-based top-N (no ranking) → sorts full match set by date.
+- Single JS chunk (~219 KB gzip) slows cold loads.
+- Full-mailbox sync latency at 100k+ (batching/Polly present, but per-message fetch).
+
+## Documentation quality
+Strong for a scaffold: `README`, `docs/ARCHITECTURE.md`, OAuth setup, `docs/specs/*`,
+`AGENTS.md`, plus `WORKFLOW.md`/`CHANGELOG`; meaningful XML-doc comments.
+**Gaps:** no API reference, no ERD/data-model doc, no ADRs, no search/AI design docs —
+which this discovery produces.
+
+## Verdict
+A clean, secure, well-tested foundation with a **genuine FTS base** and an **AI seam
+already in place**. The biggest opportunities are exactly where the product wants to
+win: **relevance-ranked, multi-mode, assisted search**; a **richer AI contract**
+(embeddings/classification/extraction); **conversation intelligence**; and a **modern,
+approachable UX**. None of these require a rewrite — they extend the existing seams.

docs/discovery/02-competitor-analysis.md

@@ -0,0 +1,73 @@
+# 02 — Competitor Analysis (Phase 2)
+
+Goal: understand the modern email landscape, isolate what users consistently praise
+and complain about, and find **where InboxIntel can be genuinely better** — without
+copying anyone. Analysis is framed against the [Design Brief](00-design-brief.md):
+approachable-professional, fast, **local/private AI**, search-centric.
+
+## Landscape snapshot
+| Product | Positioning | Standout strength | Recurring weakness |
+|---------|-------------|-------------------|--------------------|
+| **Gmail** | Default mass-market webmail | Powerful search operators; scale; free; Gemini bolt-ons | Cluttered; privacy optics; AI feels tacked-on; dated triage |
+| **Outlook** | Enterprise mail + calendar | Calendar/email fusion; Copilot; Rules | Heavy; **search is famously flaky/slow**; inconsistent "new Outlook" |
+| **Superhuman** | Speed tool for pros | **Blazing speed, keyboard triage, polish**; Superhuman AI | $30/mo; Gmail/Outlook-only; **intimidating for casual users** |
+| **Proton Mail** | Privacy / E2EE | Zero-access encryption; Swiss; open source; local-ish Scribe AI | **Search limited** (encrypted); fewer productivity features; slower |
+| **Spark** | Smart inbox + team email | Collaboration (shared drafts, comments), smart inbox, cross-platform | Privacy history concerns; can feel cluttered; sync hiccups |
+| **Shortwave** | AI-native Gmail client | **AI search + assistant that genuinely work**; bundles; fast | Gmail-only; **cloud AI (privacy)**; pricing crept up |
+| **Thunderbird** | Open-source local client | Free, private, extensible, local storage, multi-account | UX still catching up; **no built-in AI**; search not modern |
+| **HEY** | Opinionated workflow | Novel triage (Screener/Imbox/Feed/Paper Trail) | Locked-in; no folders/search-first; pricey |
+| **Fastmail** | Fast private power-mail | Excellent fast search; reliable; standards-based | No AI; power-user aesthetic; niche |
+| **Missive** | Team shared inbox | Best-in-class collaboration/assignment/rules | Team-first; overkill for individuals |
+
+## Deep dive on the two axes that decide this market
+
+### Search (InboxIntel's flagship battleground)
+- **Gmail / Fastmail**: fast, operator-rich, but **syntax-first** and browse-anchored — great for people who already know `from:` / `has:attachment`, opaque to everyone else.
+- **Outlook**: powerful on paper, but **unreliable/slow search is its most complained-about feature** for years.
+- **Shortwave**: the modern bar — **AI/semantic search + "ask your inbox"** questions ("what did Sarah say about the invoice?"). Genuinely loved, but **cloud-processed**.
+- **Proton/Thunderbird**: privacy-strong but **search is a known weak point** (encryption / dated indexing).
+- **Gap:** nobody offers **fast, relevance-ranked, semantic, *explainable* search that is also local/private and approachable to non-power-users.** That is precisely InboxIntel's opening.
+
+### AI
+- The market has **split into two camps**:
+  1. **AI, but cloud** — Gmail/Gemini, Outlook/Copilot, Shortwave, Spark. Useful, but your email is processed off-device.
+  2. **Private, but little/no AI** — Proton, Thunderbird, Fastmail.
+- Proton's **Scribe** (privacy-first, can run locally) hints at the future but is narrow (writing only).
+- Common complaints: AI feels **generic/bolted-on**, **unexplained**, and **can't be turned off cleanly**.
+- **Gap:** **genuinely useful AI that runs 100% locally (Ollama), is modular, explainable, and fully optional** — nobody mainstream owns this.
+
+## What users consistently PRAISE (cross-product)
+1. **Speed** — instant search/open/triage (Superhuman, Fastmail).
+2. **AI that saves real time** — summaries of long threads, "ask your inbox," draft replies (Shortwave, Superhuman AI).
+3. **Privacy you can trust** — Proton's whole brand.
+4. **Effortless triage** — Split Inbox / Bundles / Screener (Superhuman, Shortwave, HEY).
+5. **Collaboration** — assign, comment, share drafts (Missive, Spark).
+6. **Clean, calm, modern UI** that reduces overwhelm.
+
+## What users consistently COMPLAIN about
+1. **Search that's slow, flaky, or syntax-only** (Outlook; Gmail/Proton on mobile).
+2. **Privacy cost of AI** — "I want the AI but not to send my mail to a server."
+3. **Clutter & overwhelm** — noisy inboxes, ads, too many features.
+4. **AI that's generic and unexplained** — "why did it say that / categorise that?"
+5. **Power tools that alienate casual users** (Superhuman's learning curve; HEY's rigidity).
+6. **Price** — best experiences gated behind $10–30/mo.
+7. **Opaque automation/categorisation** you can't understand or correct.
+
+## Opportunity gaps → where InboxIntel wins (ranked)
+1. **Local, private, *useful* AI.** Resolve the market's central tension (AI **vs** privacy) by delivering both via Ollama. This is the headline differentiator and aligns with the existing `IAIProvider` seam and read-only scope.
+2. **Approachable power-search as the home screen.** Discoverable, visual, assisted (filter chips + NL + suggestions) so a *mainstream* user searches confidently — with operators/semantic power underneath for those who grow into it.
+3. **Explainability everywhere.** "Why this result matched," "why this was categorised/prioritised." No competitor does this well; it builds trust *and* teaches the product — perfect for the "anyone can pick it up" audience.
+4. **Own-your-data intelligence.** Analytics, cleanup, unsubscribe, and local AI combine into a "your inbox, your machine, your intelligence" story that Gmail/Outlook structurally can't tell.
+5. **Fast *and* friendly.** Superhuman is fast-but-intimidating; casual apps are friendly-but-slow. The brief's sweet spot (balanced, pointer-first, polished) is largely unoccupied.
+6. **Honest, correctable automation.** Categories/priority a user can see, understand, and fix — the antidote to Gmail's opaque tabs.
+
+## Explicitly NOT copying
+- Not Superhuman's keyboard-only speed-sport (brief = discoverable-first).
+- Not HEY's rigid, search-hostile opinionation.
+- Not cloud-AI-at-any-cost (brief = local/private).
+- Not enterprise-collaboration-first (individual mainstream user is primary; collaboration is a later opportunity, see roadmap).
+
+## Positioning statement (draft)
+> **InboxIntel** — the email tool that makes **search the fastest way to think about your
+> inbox**, with **AI that runs on your own machine** and always explains itself.
+> Powerful enough for pros, simple enough for anyone.

docs/discovery/03-user-research.md

@@ -0,0 +1,107 @@
+# 03 — User Research: Personas & Journey Maps (Phase 3)
+
+Personas are prioritised against the [Design Brief](00-design-brief.md) audience:
+**mainstream, power-capable.** So the **primary** personas are everyday individuals and
+small-business owners; **secondary** personas are high-volume role users whose needs we
+support *progressively* (power features revealed as people grow into them).
+
+Legend: 🟢 primary · 🔵 secondary
+
+---
+
+## 🟢 Priya — Personal user (mainstream)
+Busy professional with a personal Gmail full of subscriptions, receipts, travel, and the
+occasional important thread buried in noise.
+- **Goals:** find things fast; not miss the important stuff; keep the inbox from feeling overwhelming.
+- **Pain points:** can't remember where things are; newsletters bury real mail; unsubscribing is tedious; search needs the "right words."
+- **Daily workflow:** skim inbox on phone/desktop → star/leave a few → hunt for a receipt or booking → ignore the rest.
+- **Search needs:** *natural, forgiving* ("flight to Lisbon", "gym receipt March") — no operators. Attachment/receipt finding. Typo tolerance.
+- **AI opportunities:** thread summaries, "what needs a reply," smart unsubscribe suggestions, receipt/booking extraction, gentle priority.
+
+## 🟢 Marcus — Business owner / solopreneur (mainstream, high-value)
+Runs a small business from one inbox: clients, vendors, invoices, admin — **no assistant**.
+- **Goals:** never drop a client ball; find any past agreement/invoice instantly; spend less time in email.
+- **Pain points:** follow-ups slip; important buried under admin; digging for "what did we agree?"; context-switching.
+- **Daily workflow:** triage first thing → reply to clients → chase unpaid invoices → search for prior context mid-reply.
+- **Search needs:** people-centric ("everything with this client"), attachment/invoice search, timeline ("our thread about the contract"), "did they reply?"
+- **AI opportunities:** follow-up detection, "you haven't heard back" nudges, thread/relationship summaries, task/commitment extraction, draft replies with context.
+
+## 🔵 Dev — Developer / power user
+Technical inbox: GitHub, CI alerts, monitoring, mailing lists, plus real mail.
+- **Goals:** cut notification noise; automate; keep signal. **Strongly values local/private AI** and self-hostable.
+- **Pain points:** alert floods; wants rules/automation and keyboard speed; distrusts cloud AI on their mail.
+- **Daily workflow:** bulk-archive alerts → scan PR/issue threads → deep-search history → automate the repetitive.
+- **Search needs:** operators + regex-ish precision, entity/sender/domain, saved searches, fast keyboard-driven search.
+- **AI opportunities:** local summarisation of long threads, auto-labelling/rules suggestions, semantic search across archives, duplicate/near-duplicate detection.
+
+## 🔵 Rina — Recruiter (role power user, high volume)
+Hundreds of candidate threads, CV attachments, scheduling, multi-stage follow-ups.
+- **Goals:** find any candidate/CV instantly; track stage; never lose a promising lead.
+- **Pain points:** attachments scattered; who's at what stage; follow-up timing; duplicate applicants.
+- **Search needs:** **attachment/document search** (search *inside* CVs), people search, stage/timeline, entity extraction (skills, roles).
+- **AI opportunities:** CV/document understanding, candidate summaries, follow-up detection, duplicate detection, relationship mapping.
+
+## 🔵 Sam — Sales · 🔵 Tom — Support (role power users)
+- **Sam:** pipeline follow-ups, response tracking, templates. Needs "who hasn't replied," saved searches, sentiment, reply suggestions.
+- **Tom:** triage by category/SLA, canned responses, categorisation accuracy. Needs reliable auto-categorisation, priority prediction, phishing/spam confidence, thread summaries.
+- (Both hint at **collaboration/shared-inbox** — deliberately a *later* opportunity; individual mainstream user is primary.)
+
+---
+
+## Cross-persona signal
+| Need | Who feels it most | Priority |
+|------|-------------------|----------|
+| **Forgiving, natural search** | Priya, Marcus (everyone) | 🔴 Highest |
+| **Find people / attachments / past context** | Marcus, Rina, Dev | 🔴 High |
+| **Follow-up / "did they reply?" detection** | Marcus, Sam, Rina | 🔴 High |
+| **Thread & relationship summaries** | Marcus, Dev, Rina | 🟠 Medium-high |
+| **Noise reduction (unsubscribe, bulk, categorise)** | Priya, Dev, Tom | 🟠 Medium-high |
+| **Local/private AI** | Dev (loud), all (latent) | 🔴 Strategic |
+| **Explainability of results/priority** | everyone (trust) | 🟠 Medium-high |
+| Collaboration / shared inbox | Sam, Tom | 🟢 Later |
+
+---
+
+## Journey maps
+Format: **stage → what they do → current pain → InboxIntel opportunity.**
+
+### J1 — Morning triage ("what needs *me* today?") — Priya, Marcus
+| Stage | Action | Current pain | Opportunity |
+|-------|--------|--------------|-------------|
+| Arrive | Open inbox | Wall of mixed noise + important | AI **priority lane** + summary of "needs you" (local, explainable) |
+| Scan | Skim subjects | No signal of what's urgent/awaiting reply | **Follow-up/awaiting-reply** badges; thread one-liners |
+| Act | Reply/defer/clean | Repetitive, context-switch heavy | One-click defer/snooze; context-aware **draft reply**; bulk clean noise |
+| Exit | Close, hope nothing missed | Anxiety of missed items | "You're caught up on what matters" confidence state |
+
+### J2 — Find one specific thing (the flagship moment) — everyone
+| Stage | Action | Current pain | Opportunity |
+|-------|--------|--------------|-------------|
+| Recall | Remember fragments ("invoice, that vendor, spring") | Must guess exact keywords/operators | **Natural-language + semantic** search; typo-tolerant |
+| Query | Type in search | Syntax anxiety; blank box | **Suggestions, recent, filter chips**; search-as-home |
+| Scan results | Look through hits | Date-sorted, not relevant; no context | **Relevance ranking** + **"why this matched"** + instant preview |
+| Confirm | Open the right one | Re-open several to be sure | Grouped results, inline preview, people/attachment facets |
+
+### J3 — Reduce the noise — Priya, Dev, Tom
+| Stage | Action | Current pain | Opportunity |
+|-------|--------|--------------|-------------|
+| Notice | "Too many newsletters" | Unsub links hidden/dark-patterned | **One-click unsubscribe** (already seeded) + AI suggestions of what to cut |
+| Decide | Which to keep? | Manual, per-email | AI **bulk suggestions** by sender/category with confidence + preview |
+| Act | Unsubscribe / bulk clean | Risky, irreversible-feeling | **Confirmed + previewed** bulk actions (already a strength) |
+| Trust | Did it do the right thing? | Opaque | Explainable "why suggested," undo, audit |
+
+### J4 — Track a commitment / follow-up — Marcus, Sam, Rina
+| Stage | Action | Current pain | Opportunity |
+|-------|--------|--------------|-------------|
+| Send | Email a client/candidate | Then forget | AI **follow-up detection**: "expecting a reply?" |
+| Wait | Time passes | No system nudges you | "**No reply in N days**" surfaced automatically |
+| Recall | "What did we agree?" | Dig through thread | Thread **summary + extracted commitments/tasks** |
+| Resolve | Chase / close | Re-draft from scratch | Context-aware **draft nudge** referencing the thread |
+
+---
+
+## Research-driven product principles
+1. **Search is the front door**, and it must work for someone who types *"that gym receipt"* — not just `from:gym has:attachment`.
+2. **Explain everything** the system decides (match, category, priority) — it builds trust and teaches the app.
+3. **AI removes toil, never control** — advisory, previewed, undoable, and fully optional/local.
+4. **Progressive power** — mainstream-simple by default; operators, saved searches, automation revealed as users grow.
+5. **Reduce anxiety** — "you're caught up," undo, and confidence states matter as much as features.

docs/discovery/04-ux-redesign-and-design-system.md

@@ -0,0 +1,182 @@
+# 04 — UX/UI Redesign, Information Architecture & Design System (Phase 4A)
+
+A **complete redesign**, not an iterative tweak. Nothing about the current layout is
+assumed to survive. Governed by the [Design Brief](00-design-brief.md): Notion/Arc-
+professional, pointer-first & discoverable, balanced density, dark-first, green
+`#3ba31f`, subtle motion, fully responsive, mainstream-friendly.
+
+---
+
+## Part A — UX Research (per-screen evaluation)
+Evaluating the *jobs*, not the current pixels. For each screen: **goal · is it intuitive
+· what's unnecessary · how to simplify · click reduction · what must be obvious · sources
+of cognitive overload.**
+
+| Screen | User goal | Redesign verdict |
+|--------|-----------|------------------|
+| **Dashboard (current draggable widgets)** | Understand my inbox | Becomes the **Analytics** view, not the home. A draggable widget grid is *configuration overhead* most users never want. Default = a curated overview; customisation is progressive. |
+| **Inbox/list** | Triage what matters | Reframe from "all mail newest-first" to **lanes** (Needs you · Awaiting reply · Everything). Obvious: who/subject/one-line intent/time. Overload source: undifferentiated noise → fix with grouping + priority. |
+| **Reading a message** | Understand + act | Add a **thread summary** header, **extracted actions/dates**, and inline reply. Reduce clicks: reply/snooze/label as one-key or one-click from the pane. |
+| **Search** | Find a specific thing | **Promote to the front door.** Today it's an operator box; redesign to inviting, visual, assisted (see [05](05-search-redesign.md)). |
+| **Cleanup / unsubscribe** | Reduce noise safely | Strong bones (confirm+preview). Make it **suggestion-led** ("cut these 12 newsletters?") with confidence + undo. |
+| **Settings** | Configure incl. AI | Add a clear **AI panel**: off / local (Ollama) / provider, model status, VRAM. AI-off must feel first-class, not degraded. |
+
+**Cross-cutting UX principles:** search-as-home · lanes over one big list · explain every
+decision · progressive disclosure of power · confidence/undo everywhere · one primary
+action per screen.
+
+---
+
+## Part B — Information Architecture
+
+### Navigation shell (responsive 3-pane → 1-pane)
+```
+┌───────────────────────────────────────────────────────────────┐
+│  TopBar:  [ 🔍  Search your inbox…            ⌘K ]   ☾  ⚙  👤  │
+├──────────┬──────────────────────────┬─────────────────────────┤
+│ Sidebar  │  List / Results          │  Reading / Preview      │
+│ (collaps)│  (virtualised)           │  (thread + AI summary)  │
+│          │                          │                         │
+│ Search   │  ▸ Needs you (lane)      │  Subject                │
+│ Priority │  ▸ Awaiting reply        │  ⟶ AI summary (local)   │
+│ Unread   │  ▸ Everything            │  ⟶ extracted actions    │
+│ Saved ★  │                          │  body …                 │
+│ Categories                          │  [Reply] [Snooze] […]   │
+│ Cleanup  │                          │                         │
+│ Analytics│                          │                         │
+│ ─────────│                          │                         │
+│ 👤 acct  │                          │                         │
+└──────────┴──────────────────────────┴─────────────────────────┘
+```
+- **Search sits at the top of everything** (top bar) *and* as the first sidebar item — reinforcing search-as-home.
+- **Responsive collapse:** 3-pane (wide desktop) → 2-pane (list+reading, laptop) → 1-pane with push navigation (tablet/mobile). Reading opens as an overlay sheet on mobile.
+- **Progressive disclosure:** advanced filters, operators, saved-search management, and automation are revealed on demand — never in a beginner's face.
+- **Command palette (⌘K / Ctrl+K):** an *accelerator* — navigate, act, and search — layered on top of the fully clickable UI (pointer-first per brief).
+- **Right-click context menus** on rows/senders/threads (archive, label, unsubscribe, "find similar," "everything from this sender").
+- **Multi-window / dockable panels:** deferred (web app); pop-out reading view is a v-later opportunity.
+
+### Screen hierarchy
+1. **Search-home** (front door) · 2. **List/Results** (lanes, ranked) · 3. **Reading/thread**
+· 4. **Cleanup** · 5. **Analytics** · 6. **Settings (incl. AI)**.
+
+---
+
+## Part C — Design System
+
+### Colour — accent ramp (from `#3ba31f`)
+```
+green-50  #f890? →  use tint set:
+--green-50:  #f1f9ec   --green-300: #93d07d   --green-600: #2f8419
+--green-100: #dcf0d0   --green-400: #63b84a   --green-700: #266a15
+--green-200: #bde3ab   --green-500: #3ba31f   --green-800: #1e5312
+                        (brand base)          --green-900: #163a0e
+```
+**Usage rules (dark-first):**
+- Brand/base = `green-500`. On dark surfaces, interactive fills use `green-500`/`green-400`; **foreground on accent is contrast-checked** (near-black `#0f1a0b` on light greens, white on `green-600`+).
+- Accent is used **sparingly** — primary actions, selection, active nav, positive status.
+- **Never colour-only:** selection also shows a left-bar/checkbox; status pairs green with an icon/label (colour-blind-safe by construction, even though formal a11y is deferred).
+
+### Colour — neutrals (warm-leaning)
+| Token | Dark (default) | Light |
+|-------|----------------|-------|
+| `--bg` | `#1a1917` (warm charcoal, **not** pure black) | `#ffffff` (genuine white) |
+| `--surface-1` | `#211f1d` | `#faf9f7` |
+| `--surface-2` | `#2a2724` | `#f4f2ee` |
+| `--surface-3` | `#34302c` | `#ebe8e2` |
+| `--border` | `#3a3632` | `#e4e0d9` |
+| `--text` | `#f2efe9` (warm off-white) | `#1c1a17` |
+| `--text-muted` | `#a8a29a` | `#6b6459` |
+| `--text-subtle` | `#7a746c` | `#928b7e` |
+| semantic | `info #4a90d9 · warn #d9a441 · danger #d95a4a · success = green-500` | same, contrast-tuned |
+
+### Typography
+- **UI font:** Inter (or system fallback) — clean, neutral, highly legible.
+- **Optional mono:** JetBrains Mono / ui-monospace for addresses, IDs, data.
+- **Scale (px / line-height), UI base = 14 for balanced density:**
+  `xs 12/16 · sm 13/18 · base 14/20 · md 16/24 · lg 18/26 · xl 20/28 · 2xl 24/32 · 3xl 30/38`
+- Weights: 400 body · 500 UI/labels · 600 headings/emphasis. Avoid 700 except brand.
+
+### Spacing (4px base) & layout
+`space: 2, 4, 6, 8, 12, 16, 20, 24, 32, 40, 48, 64`.
+Grid: 12-col fluid content area; sidebar fixed (240px, collapsible to 56px icon rail);
+reading pane min 420px. Density "balanced" → row height ~44px, 8–12px internal padding.
+
+### Radius / elevation / motion
+- **Radius:** `sm 4 · md 6 (buttons/inputs) · lg 8 (cards/panels, default) · xl 12 (modals) · pill 999`.
+- **Elevation:** dark = surface-layering + faint shadow + 1px border; light = soft shadows
+  `e1 0 1 2 /6% · e2 0 4 12 /10% · e3 0 12 32 /16%`. Levels: e0 flat · e1 menus · e2 popovers · e3 modals.
+- **Motion:** durations `120 / 180 / 240ms`; easing `cubic-bezier(0.2,0,0,1)` (ease-out) for enters, `cubic-bezier(0.4,0,1,1)` for exits; a spring only for selection/drag. **Respect `prefers-reduced-motion`.**
+
+### Icons & illustration
+- **Lucide** (line, rounded), stroke 1.5px, 20px default (16px dense, 24px feature).
+- Illustration: minimal, single-accent line spot-art for empty states — friendly, not corporate stock.
+
+### States (must all be designed)
+- **Loading:** skeleton rows (list), shimmer summary card (reading) — never spinners for content.
+- **Empty:** friendly line-art + one clear CTA ("Nothing here yet — connect Gmail" / "No results — try broader terms" with a *Did you mean* / *Broaden* action).
+- **Error:** calm, specific, recoverable ("Couldn't reach Gmail — Retry"), never a raw stack.
+- **Success:** toast + inline confirmation; destructive actions show **preview → confirm → undo**.
+
+### Components (catalogue)
+Buttons (`primary` green / `secondary` surface / `ghost` / `danger`) · icon-button · input
+· **search field** (hero variant) · **filter chip** (removable, typed) · segmented control ·
+toggle · dropdown menu · **context menu** · **command palette** · **email row** (avatar,
+sender, subject, one-line intent, badges, time, hover-actions) · **sender chip/avatar** ·
+**thread summary card** · **category badge** · **priority indicator** · tabs · tooltip ·
+toast · modal · **side sheet** (mobile reading) · skeletons · empty-state · pagination /
+**virtualised infinite scroll** · avatar/initials · progress/VRAM meter (AI panel).
+
+---
+
+## Part D — Themes (both fully polished)
+
+### Dark (default)
+Layered **warm charcoal** surfaces (`#1a1917` → `#34302c`), warm off-white text, green
+accent nudged for on-dark contrast, faint borders to separate layers. Avoids pure black;
+depth via surface elevation + hairline borders, not heavy shadow.
+
+### Light
+**Genuine white** base (`#ffffff`) with warm off-white surfaces — *not grey-pretending-to-
+be-white*. Generous whitespace, soft shadows for elevation, restrained green accent.
+Premium, low-noise, highly readable.
+
+Both share tokens; only the neutral map + shadow strategy differ. Theme follows a
+`data-theme` attribute; **dark is the design source of truth**, light is derived and
+independently QA'd.
+
+---
+
+## Part E — Interaction design
+- **Hover:** rows raise to `surface-2`, reveal quick-actions (archive/snooze/label/unsub).
+- **Selection:** checkbox on hover + click-row-to-open; shift/⌘-click multi-select; a
+  sticky **bulk action bar** slides up when >1 selected.
+- **Search:** instant results, **live filtering** as chips are added/removed, suggestions +
+  recent on focus (see [05](05-search-redesign.md)).
+- **Previews:** hover peek + inline reading; attachments preview in a lightbox.
+- **Drag & drop:** rows → labels/categories/cleanup; respects reduced-motion.
+- **Notifications:** toasts (non-blocking), with undo for reversible actions.
+- **Transitions:** pane content crossfades; mobile reading slides up as a sheet.
+- **Scrolling:** **virtualised list** (mandatory for 100k+ rows) with sticky lane headers.
+
+---
+
+## Part F — Accessibility review (baseline; deeper a11y deferred)
+Baked in cheaply now: AA-tuned contrast via the ramps, visible focus rings, `prefers-
+reduced-motion`, **non-colour-only** state, semantic HTML + ARIA on lists/dialogs/menus,
+full keyboard reachability of core actions. **Deferred to pre-launch backlog** (per brief):
+high-contrast mode, formal screen-reader passes, font-scaling controls, colour-blind audit.
+
+---
+
+## Part G — Migration strategy (strangler, low-risk)
+1. **Introduce tokens + component library** (Tailwind config from this doc) — no behaviour change.
+2. **Rebuild the shell** (sidebar / top-bar / search-home) around the existing API.
+3. **Migrate screen-by-screen behind a feature flag** (`ui.v2`): Search → Reading → List →
+   Cleanup → Analytics (the old draggable dashboard *becomes* Analytics).
+4. **Keep the API stable**; frontend-only migration. Delete old screens once parity is verified.
+5. Ship per-screen via the `develop → staging` pipeline; each screen is its own epic (see [Git plan](10-git-plan.md)).
+
+## Part H — Future design opportunities
+Pop-out / multi-window reading · dockable panels · custom accent picker · additional themes
+· a Tauri/Electron shell if a true desktop build is ever wanted · plugin-contributed widgets
+on the Analytics canvas · command-palette extensibility.

docs/discovery/05-search-redesign.md

@@ -0,0 +1,129 @@
+# 05 — Search Redesign (Phase 4B)
+
+Search is the **flagship**. The goal: make people *want to search rather than browse* —
+an experience that's **inviting, visual, forgiving, ranked, and explainable**, works for
+*"that gym receipt from March"* (not just `from:gym has:attachment`), and runs **locally
+and privately**. Grounded in the existing Postgres FTS + `IAiProvider` seam (see
+[01](01-architecture-review.md)); AI modes are **optional** — with AI off, lexical +
+structured search is still excellent.
+
+---
+
+## The layered search engine
+Four cooperating layers, each usable alone; higher layers *degrade gracefully* to lower
+ones when AI is disabled.
+
+```
+┌ Layer 4 · AI-ASSISTED  (optional, local) ────────────────────────────┐
+│  NL→query · conversational "ask your inbox" · explanations · related   │
+├ Layer 3 · SEMANTIC  (optional, local embeddings + pgvector) ──────────┤
+│  meaning-based recall · "similar to this" · concept search             │
+├ Layer 2 · LEXICAL + RANKING  (Postgres FTS, always on) ───────────────┤
+│  websearch_to_tsquery · ts_rank · pg_trgm fuzzy · multi-field          │
+├ Layer 1 · STRUCTURED  (SQL WHERE, always on) ─────────────────────────┤
+│  sender/date/labels/flags/size/category · filter chips                 │
+└───────────────────────────────────────────────────────────────────────┘
+                 ▲ hybrid fusion (RRF) ranks Layer 2+3 together ▲
+```
+
+**Hybrid ranking (RRF):** lexical and semantic candidate sets are merged via *Reciprocal
+Rank Fusion*, then boosted by **recency**, **sender importance**, and **unread**. This
+replaces today's date-only ordering — the single biggest search win.
+
+---
+
+## The experience: search as the front door
+- **Hero search field** in the top bar + a **search-home** when the box is focused/empty:
+  **recent searches**, **suggested searches** ("Unread from people you reply to",
+  "Large attachments", "Receipts this month"), and **saved searches ★**.
+- Type freely → **live results** with **filter chips** the user can add/remove by click
+  (pointer-first). Behind an unobtrusive "＋ Filters" for the full builder.
+- Every result shows **"why it matched"** (highlighted terms, or the semantic concept, or
+  the parsed NL interpretation as *editable chips*).
+- **Zero-result recovery:** *Did you mean* · *Broaden* · *Search all mail/trash*.
+- Fully keyboard-drivable as an **accelerator** (⌘K → type → arrow → enter), but never required.
+
+---
+
+## Search modes & features
+Each with **Problem solved · Implementation · Complexity · Effort · User value.**
+Effort = rough dev-days for one engineer; Complexity = S/M/L/XL.
+
+| Feature | Problem solved | Implementation | Cx | Effort | Value |
+|---------|----------------|----------------|----|--------|-------|
+| **Relevance ranking** | Results sorted by date, not usefulness | `ts_rank_cd` + hybrid RRF + recency/sender/unread boosts; top-N by score | M | 3–5d | 🔴🔴🔴 |
+| **Multi-field FTS** | Only Subject+Body indexed | Extend generated `tsvector` (sender name/addr, attachment filenames, labels) with weights (A/B/C/D) | S | 2–3d | 🔴🔴 |
+| **Fuzzy / typo tolerance** | "recieved" finds nothing | `pg_trgm` GIN + similarity fallback when FTS is empty | S | 1–2d | 🔴🔴 |
+| **Phrase / prefix / boolean** | `plainto_tsquery` too blunt | Swap to `websearch_to_tsquery` (quotes, OR, -exclude) | S | 1d | 🔴 |
+| **Interactive filter chips** | Operators are a syntax wall | UI chips ↔ structured DTO; add/remove live | M | 3–4d | 🔴🔴🔴 |
+| **Advanced query builder** | Power users want precision | Visual builder → same DTO; operators still work | M | 3–4d | 🔵🔵 |
+| **Natural-language search** | "gym receipt March" | Local LLM (or rules-first) parses intent → chips + terms, shown editable | L | 5–8d | 🔴🔴🔴 |
+| **Semantic search** | Recall by *meaning*, not keywords | `IEmbeddingProvider` (Ollama) → `pgvector` (HNSW); backfill job; hybrid with FTS | L | 8–12d | 🔴🔴🔴 |
+| **Conversational "ask your inbox"** | "What did Sarah say about the invoice?" | RAG: semantic retrieve → local LLM answers with **citations** to source emails | XL | 10–15d | 🔴🔴🔴 |
+| **"Why this matched"** | Trust + learnability | `ts_headline` highlights (lexical); nearest-concept + snippet (semantic); parsed-intent chips (NL) | M | 3–4d | 🔴🔴 |
+| **Saved searches ★** | Repeated queries | Persist DTO; pin to sidebar; optional live count | S | 2d | 🔴🔴 |
+| **Recent searches / history** | Re-find what you searched | Per-user history table; privacy-clearable | S | 1–2d | 🔴 |
+| **Suggested searches** | Blank-box paralysis | Rules over user's own stats (top senders, unreplied, big attachments) | S | 2–3d | 🔴🔴 |
+| **People search** | "Everything with this person" | Sender/recipient facet + aggregation; person profile panel | M | 3–5d | 🔴🔴 |
+| **Attachment search** | Find files, not emails | Index filenames now; **content/OCR** later (see below) | S→L | 2d → +8d | 🔴🔴 |
+| **Entity extraction** | Search by amounts/dates/orgs | Local NER (LLM or rules) → entity index; facets (money, dates, companies) | L | 6–10d | 🔴🔴 |
+| **Timeline search** | "our thread over time" | Date-bucketed results + a timeline scrubber UI | M | 3–4d | 🔵🔵 |
+| **Related conversations** | Surface adjacent context | Vector nearest-neighbours + same-participants/thread heuristics | M | 3–5d | 🔴🔴 |
+| **Duplicate / near-dup detection** | Clutter, repeated sends | Hash (exact) + embedding similarity (near) → group/cleanup | M | 4–6d | 🔵🔵 |
+| **Conversation intelligence** | Long threads are walls | Thread summary + extracted actions/decisions (local LLM) surfaced in results & reading | L | 8–12d | 🔴🔴🔴 |
+
+### Additional ideas (beyond the brief)
+- **Search-driven bulk actions** — run cleanup/label/unsub on a *result set* ("archive all newsletters older than 6 months").
+- **Scoped search** — constrain to this folder / label / person / thread inline.
+- **"Find similar to this email"** — one-click semantic neighbours from any message.
+- **Zero-result recovery** — *did-you-mean* (trigram), *broaden*, *search trash/all*.
+- **Search from selection** — highlight text → "search inbox for this".
+- **Sender reputation / safety facet** — filter by phishing/spam confidence (ties to AI).
+- **Search templates** — parameterised saved searches ("invoices from {client}").
+- **Search analytics (for the user)** — "you search 'invoice' most" → suggests a saved search / smart folder.
+
+---
+
+## Ranking design (detail)
+```
+score = RRF(lexical_rank, semantic_rank)
+        + w_recency   · decay(sentAt)
+        + w_sender    · senderImportance(userReplyRate, frequency)
+        + w_unread    · isUnread
+```
+- Weights are configurable; defaults tuned so **exact/lexical hits never lose to fuzzy noise**.
+- `senderImportance` derives from the user's *own* behaviour (reply rate, frequency) —
+  computable from existing data, no AI required.
+
+## Performance
+- **Indexes:** GIN on `SearchVector`; GIN `pg_trgm` on sender/subject; **HNSW** (or IVFFlat)
+  on the `pgvector` embedding column.
+- **Keyset/cursor pagination** for lexical/structured (replaces offset `Skip/Take`); score-
+  ordered top-N for ranked/semantic (fetch N, no deep offset).
+- **Async embedding backfill** worker (reuse the hosted-worker pattern) so semantic search
+  builds in the background; new mail embedded on sync. Batch to respect the 3080's VRAM.
+- Cache suggestions/recent; debounce live search (~120ms).
+- Target: **<150ms** lexical, **<400ms** hybrid on 100k emails (local).
+
+## Privacy
+- Embeddings and NER computed **locally via Ollama**, stored **locally** in Postgres.
+- Nothing leaves the machine unless the user explicitly selects a cloud provider.
+- **AI-off fallback:** Layers 1–2 (structured + ranked lexical + fuzzy) — still a top-tier
+  search. Semantic/NL/conversational simply hide when AI is disabled.
+
+## Mapping to the existing code (extend, don't rewrite)
+- **`GmailQueryParser` → `SearchIntentParser`:** keep operator parsing; add a rules-first
+  NL layer, optional local-LLM disambiguation, emit the same `SearchRequestDto` (+ new fields).
+- **`SearchService`:** add ranking (`ts_rank_cd`), `websearch_to_tsquery`, `pg_trgm`
+  fallback, multi-field vector, hybrid RRF, keyset pagination, `ts_headline` explanations.
+- **`Email.SearchVector`:** widen the generated column (weighted A/B/C/D) + add a
+  `pgvector` `Embedding` column + `EmailEntities`/`ThreadSummary` tables.
+- **AI seam:** extend to `IEmbeddingProvider.EmbedAsync` and a structured
+  `IAiProvider.CompleteStructuredAsync` (see [06 AI Strategy](06-ai-strategy.md)); Null
+  provider returns "unavailable" so callers fall back to lexical.
+
+## Phasing (feeds the roadmap)
+- **MVP search:** relevance ranking · multi-field FTS · fuzzy · chips · saved/recent/suggested · "why matched" (lexical) · keyset pagination.
+- **v1.1:** natural-language parse · people search · attachment (filename) · search-driven bulk.
+- **v1.2:** semantic (pgvector) · related · find-similar · thread summaries in results.
+- **v2.0:** conversational "ask your inbox" (RAG + citations) · entity search · duplicate detection · attachment OCR/content.

docs/discovery/06-ai-strategy.md

@@ -0,0 +1,113 @@
+# 06 — AI Strategy & Model Recommendations (Phase 5)
+
+AI is **optional, modular, local-first, and explainable**. The app must be fully usable
+with AI disabled. Primary hardware: **NVIDIA RTX 3080 (10 GB VRAM)**. All AI sits behind
+an abstraction so models/providers swap without touching application logic.
+
+---
+
+## Part A — The AI abstraction (extends the existing seam)
+Today there is one interface, `IAiProvider.CompleteAsync(system, user)`, with
+`NullAiProvider` / `OllamaProvider` / `OpenAiProvider`. We extend it into a small set of
+capability interfaces plus a task-oriented facade.
+
+### Provider-level interfaces (Infrastructure)
+```csharp
+public interface IAiProvider {                    // chat / generation (exists)
+    AiProviderMode Mode { get; }
+    Task<string> CompleteAsync(string system, string user, CancellationToken ct = default);
+    // NEW: schema-constrained JSON (Ollama `format`, OpenAI `response_format`)
+    Task<T?> CompleteStructuredAsync<T>(string system, string user, CancellationToken ct = default);
+}
+
+public interface IEmbeddingProvider {             // NEW: vectors for semantic search/dedup
+    Task<float[]> EmbedAsync(string text, CancellationToken ct = default);
+    Task<IReadOnlyList<float[]>> EmbedBatchAsync(IReadOnlyList<string> texts, CancellationToken ct = default);
+}
+
+public interface IAiCapabilities {                // NEW: lets the UI hide what isn't available
+    bool Chat { get; } bool Embeddings { get; } bool Vision { get; } bool StructuredJson { get; }
+}
+```
+`NullAiProvider`/`NullEmbeddingProvider` return empty/"unavailable" so **every caller has a
+graceful non-AI path** (fall back to lexical search, heuristics, or hide the feature).
+
+### Task facade (Application) — the only thing feature code calls
+```csharp
+public interface IInboxAi {
+    Task<AiResult<string>>        SummarizeThreadAsync(Guid threadId, CancellationToken ct);
+    Task<AiResult<Extraction>>    ExtractAsync(string body, ExtractKinds kinds, CancellationToken ct);
+    Task<AiResult<Classification>>ClassifyAsync(string subject, string body, CancellationToken ct);
+    Task<AiResult<Answer>>        AskAsync(string question, SearchScope scope, CancellationToken ct); // RAG
+    Task<float[]?>                EmbedAsync(string text, CancellationToken ct);
+    Task<AiResult<RiskVerdict>>   AssessPhishingAsync(EmailContext ctx, CancellationToken ct);
+}
+```
+- `AiResult<T>` carries `{ value, available, model, latencyMs, explanation }` → powers the
+  **"explainability"** and graceful-degradation requirements everywhere.
+- A **model router** maps *logical task → model name* from config, so swapping a model is a
+  config change: `Ai:Models:{Chat,Summarize,Embed,Classify,Extract,Vision}`.
+- Prompt templates are **versioned files**, separate from code.
+- Cross-cutting: async, `CancellationToken`, per-call timeout + **Polly** fallback to the
+  Null path, and a token/VRAM-aware concurrency limiter.
+
+### Config shape
+```
+Ai:Mode = Disabled | LocalOllama | Cloud...
+Ai:Ollama:BaseUrl, KeepAliveSeconds, VramBudgetMb
+Ai:Models:Chat = qwen2.5:7b-instruct
+Ai:Models:Embed = nomic-embed-text
+Ai:Models:Vision = qwen2.5vl:7b        (load-on-demand)
+```
+
+---
+
+## Part B — Model recommendations for the RTX 3080 (10 GB)
+
+**Budget reality:** a 7–8B instruct model at Q4/Q5 (~5 GB + ~1 GB KV cache at 8k context)
+runs **alongside** a small embedding model (~0.5 GB) inside 10 GB with headroom. 14B is
+possible (~9 GB) but leaves no room for concurrency and is slower — **7–8B is the sweet spot.**
+
+| Task | Recommended (primary) | Alt | ~VRAM (Q4) | Latency | License | Notes |
+|------|----------------------|-----|-----------|---------|---------|-------|
+| **Chat / reasoning / RAG answers** | **Qwen2.5-7B-Instruct** | Llama-3.1-8B, Gemma-2-9B | ~5.5 GB | ~30–60 tok/s | Apache-2.0 | Strong reasoning, multilingual, great JSON |
+| **Summarisation** | reuse **Qwen2.5-7B** | Llama-3.1-8B | (shared) | fast | Apache-2.0 | No separate model needed |
+| **Extraction (tasks/dates/entities)** | reuse **Qwen2.5-7B** + `format:json` | Phi-4 | (shared) | fast | Apache-2.0 | Schema-constrained JSON output |
+| **Classification / tagging** | **Qwen2.5-3B** (fast) *or* embeddings-zero-shot | rules-first (exists) | ~2.2 GB | very fast | Apache-2.0/Qwen | High-volume → small model or embeddings, not 7B |
+| **Embeddings (semantic search, dedup, zero-shot class.)** | **nomic-embed-text** (768-d, 8k ctx) | mxbai-embed-large, **bge-m3** (multilingual) | ~0.5 GB | very fast | Apache-2.0 | **Keep permanently loaded** |
+| **OCR / attachment vision** | **Qwen2.5-VL-7B** *or* **MiniCPM-V** | llama3.2-vision-11B, moondream (tiny) | ~7 GB | slow | Apache/varied | **Load-on-demand**; evicts the text LLM |
+| **Language detection** | **library, not an LLM** (fastText-lid / CLD3) | — | ~0 | instant | MIT/Apache | Don't waste VRAM on this |
+| **Translation** | reuse **Qwen2.5-7B** (multilingual) | NLLB-200 (dedicated MT) | (shared) | med | Apache-2.0 | Good for common pairs; on-demand |
+| **Spam detection** | **rules + small classifier/embeddings** | 3B LLM for edge cases | ~0–2 GB | fast | — | Traditional-first; LLM only for ambiguity |
+| **Phishing detection** | **rules/URL analysis + embeddings**, LLM **reasoning** for suspicious | Qwen2.5-7B for explanation | shared | on-demand | — | LLM explains *why*; runs async on flagged mail |
+| **Duplicate detection** | **hashing (exact) + embedding cosine (near)** | — | (uses embed) | fast | — | No chat model required |
+
+**Selection rationale:** accuracy + **Apache-2.0 licensing** (commercial-safe) + fits 10 GB
++ strong **structured-JSON** and multilingual (mailboxes aren't English-only). Qwen2.5-7B is
+a single versatile workhorse for chat/summarise/extract/translate; nomic-embed-text is a
+tiny always-on retrieval engine; a vision model is a heavy, rare, on-demand guest.
+
+### Load policy (VRAM management)
+| State | Models | Rationale |
+|-------|--------|-----------|
+| **Hot (permanent)** | `nomic-embed-text` (~0.5 GB) | Used constantly (search, dedup, classify, backfill) — must be instant |
+| **Warm (keep-alive)** | `qwen2.5:7b-instruct` (~5.5 GB) | Loaded on first chat/summarise, kept warm ~5–10 min idle via Ollama `keep_alive`, then unloaded |
+| **Cold (on-demand)** | `qwen2.5vl:7b` vision, `nllb` translation | Loaded only for the specific job; unloaded after (would otherwise exceed budget alongside the 7B) |
+- Ollama handles load/unload; we set `keep_alive` per task and a **VRAM budget guard** that
+  serialises a vision job behind unloading the text LLM.
+- Embedding **backfill** runs as a hosted background worker, batched, low priority, so it
+  never starves interactive requests.
+
+---
+
+## Part C — Principles
+1. **Traditional-first.** If rules/heuristics/library solve it well (language detection,
+   exact dedup, basic spam, most categorisation), use them — AI only where it *clearly* wins.
+2. **Always optional.** Null providers + capability flags → the app is whole without AI.
+3. **Local & private by default.** Ollama on-device; data never leaves unless a cloud
+   provider is explicitly chosen.
+4. **Explainable.** Every AI output carries a short rationale + the model used.
+5. **Swappable.** Logical-task→model routing via config; prompts versioned; no feature code
+   references a model name.
+6. **Bounded.** Timeouts, cancellation, Polly fallback, VRAM-aware concurrency — AI can never
+   hang or crash the core experience.

docs/discovery/07-ai-feature-catalogue.md

@@ -0,0 +1,70 @@
+# 07 — AI Feature Catalogue (Phase 6)
+
+Every realistic AI feature, judged honestly against the guiding rule: **AI only where it
+clearly beats traditional approaches.** Each entry: *problem · AI role & why · could
+traditional solve it? · complexity · performance · privacy.* All AI is local (Ollama),
+optional, advisory, and explainable (see [06](06-ai-strategy.md)). Complexity = S/M/L/XL.
+
+Legend for the verdict column:
+🟢 **AI clearly wins** · 🟡 **traditional-first, AI for the hard tail** · ⚪ **not AI at all**
+
+---
+
+## Group 1 — 🟢 AI clearly wins
+LLM/embeddings are genuinely the right tool; traditional approaches are weak here.
+
+| Feature | Problem | Why AI (and why traditional falls short) | Cx | Perf | Privacy |
+|---------|---------|------------------------------------------|----|------|---------|
+| **Thread / conversation summary** | Long threads are walls of text | LLMs summarise free text; regex/extractive summaries miss nuance & context | L | Warm 7B; cache per-thread, invalidate on new msg | Body → local LLM only |
+| **Conversational "ask your inbox"** (RAG) | "What did Sarah say about the invoice?" | Retrieval + generation over many emails; impossible with filters alone | XL | Semantic retrieve → 7B answer w/ **citations**; ~1–3s | Retrieval + gen fully local |
+| **Reply suggestions / writing assistant** | Blank-page drafting, tone | LLM drafts context-aware replies; templates can't adapt to content | L | Warm 7B, streamed | Thread context → local |
+| **Task / meeting / calendar / reminder extraction** | Commitments hide in prose | LLM structured-JSON extraction of {task, date, attendee}; regex catches only rigid formats | L | 7B `format:json`, async on read/sync | Body → local |
+| **Entity extraction** (amounts, orgs, dates, order #s) | Can't search/facet by meaning | LLM/NER generalises across phrasings; regex is brittle per-vendor | L | 7B or small NER, batched at sync | Local |
+| **Document / attachment understanding** | Can't search *inside* files | OCR/vision + summarise; no traditional equivalent for images/PDF meaning | XL | Vision model **on-demand** (heavy); OCR async | File content → local |
+| **Cross-thread linking / related conversations** | Related context is scattered | Embedding nearest-neighbours find semantic links; keyword join misses paraphrase | M | pgvector HNSW; precomputed | Vectors local |
+| **Relationship mapping / knowledge graph** | No view of who/what connects | Extraction + embeddings build a people/topic graph; not expressible in SQL alone | XL | Batch build; incremental | Local graph store |
+| **Conversation insights** (decisions, sentiment shift) | "What was decided / how's this going?" | LLM reads intent/sentiment over a thread; rules can't | L | 7B; cache | Local |
+| **Sentiment analysis** | Gauge tone (angry client?) | Small model/LLM classifies tone; lexicon methods are crude/misleading | M | small model or embeddings | Local |
+| **Email comparison** ("what changed vs last quote?") | Manual diffing of prose | LLM semantic diff; text-diff shows characters, not meaning | M | 7B on two bodies | Local |
+| **Explain search results / decisions** | Trust & learnability | For semantic/NL, only the model can say *why*; lexical uses `ts_headline` (non-AI) | M | cheap (reuse retrieval) | Local |
+
+## Group 2 — 🟡 Traditional-first, AI for the hard tail
+Heuristics/rules do 70–90% cheaply and instantly; AI handles ambiguity and adds explanations.
+The existing `HeuristicClassifier` and unsubscribe signals are the traditional backbone.
+
+| Feature | Problem | Traditional core | Where AI adds value | Cx | Perf / Privacy |
+|---------|---------|------------------|---------------------|----|----------------|
+| **Automatic categorisation** | Sort inbox into buckets | Rules on sender/domain/headers (exists) | Embedding zero-shot / small-LLM for the ambiguous long tail + confidence | M | Rules instant; LLM only on "unknown"; local |
+| **Smart filing / smart labels** | Where should this go? | Rules + user's past filing patterns | LLM/embedding *suggestions* with confidence, user-correctable | M | Suggest async; local |
+| **Priority prediction** | What needs me now? | Behavioural signals: your reply-rate to sender, frequency, VIPs, keywords, direct-to-me | ML/LLM refines ranking for edge cases | M | Mostly SQL/heuristic; local |
+| **Follow-up / awaiting-reply detection** | Dropped balls | Heuristic: *you* sent, contains a question, no reply in N days | LLM confirms "expects a reply" & drafts nudge | M | Heuristic instant; LLM optional; local |
+| **Smart notifications** | Notification fatigue | Rules over priority + quiet hours | LLM tunes "is this actually urgent" for borderline | S | Rules-first; local |
+| **Spam detection** | Junk | Rules/Bayesian + provider signals | Small model for novel spam; LLM explains | M | Fast; local |
+| **Phishing detection** | Safety | URL/domain analysis, SPF/DKIM hints, lookalike detection (+ existing SSRF guard) | **LLM reasons about social-engineering cues**; runs async on flagged mail, explains risk | L | Rules sync; LLM async on suspicious; local |
+| **Duplicate email detection** | Clutter / repeats | **Exact hash** for identical | **Embedding cosine** for near-duplicates | M | hash instant; vector cheap; local |
+| **Inbox assistant** (daily brief) | "Catch me up" | Compose from priority/follow-up/counts (rules) | LLM writes the natural-language brief over that structured data | L | 7B once/session; local |
+
+## Group 3 — ⚪ Not AI (don't waste VRAM)
+| Feature | Do it with | Why not AI |
+|---------|-----------|------------|
+| **Language detection** | fastText-lid / CLD3 library | Instant, accurate, ~0 VRAM; an LLM is pure overhead |
+| **Exact duplicate detection** | content hash | Deterministic and free |
+| **Unsubscribe detection** | List-Unsubscribe header parsing (exists) | Structured signal already present |
+
+---
+
+## Selection guidance (feeds the roadmap)
+- **First AI wins (highest value / lowest risk):** thread summary · follow-up detection
+  (heuristic + AI confirm) · reply suggestions · NL search parse. All reuse the one warm 7B.
+- **Semantic tier (needs pgvector + embeddings):** related/find-similar · near-dup ·
+  conversation insights · categorisation long-tail.
+- **Ambitious tier:** ask-your-inbox (RAG) · knowledge graph · attachment/vision understanding.
+- **Never gate the core on any of these** — each has a non-AI fallback or simply hides when
+  AI is off.
+
+## Privacy posture (applies to all)
+Email bodies and attachments are processed **on-device via Ollama**; embeddings, summaries,
+extractions, and graphs are **stored locally in Postgres**. No content leaves the machine
+unless the user deliberately configures a cloud provider — and even then, per-feature
+consent should gate it. This is the product's defining trust advantage (see
+[02](02-competitor-analysis.md)).

docs/discovery/08-technical-architecture.md

@@ -0,0 +1,115 @@
+# 08 — Technical Architecture (Phase 7)
+
+How the redesign is built on the **existing Clean Architecture** — extending seams, not
+rewriting. Everything here preserves the dependency rule `Api → Infrastructure →
+Application → Domain` and the security posture from [01](01-architecture-review.md).
+
+> **Update — multi-provider platform.** This architecture now assumes the
+> [multi-provider design](multi-provider/README.md): the Gmail-centric `Email`/`Sender`
+> model generalises to **account-scoped, provider-normalised** entities behind
+> `IEmailProvider`; the app becomes **small-team multi-user** (Admin/Member) with
+> **OAuth-as-login**, settings, and feature flags; and **AI gating is feature-flag-driven**
+> (`ai.enabled` system flag → user pref). See that folder for the provider abstraction,
+> DB schema, security model, and admin/settings design; this doc remains the search/AI/
+> caching/indexing reference they build on.
+
+## Overall
+```
+React SPA (v2 shell)
+   │  REST (+ SSE for streaming AI/search)
+Api  ──────────────────────────────────────────────
+Application:  IInboxAi · ISearchService · SearchIntentParser · DTOs
+Infrastructure:
+   ├ Search:   SearchService (structured + lexical rank + hybrid)
+   ├ AI:       IAiProvider / IEmbeddingProvider / model router / VRAM guard
+   ├ Jobs:     EmbeddingBackfill · AiEnrichment · IndexMaintenance (IHostedService + Channel queue)
+   ├ Gmail/Sync/Cleanup/Analytics (existing)
+   └ Persistence: EF Core + Npgsql (+ pgvector)
+Domain: entities/enums (+ Embedding, Entities, ThreadSummary, SavedSearch)
+External (optional): Ollama (local, own container) · cloud AI (opt-in only)
+```
+
+## Search architecture
+The 4-layer engine from [05](05-search-redesign.md):
+- **Structured** (SQL WHERE) + **Lexical** (`websearch_to_tsquery` + `ts_rank_cd`, weighted
+  multi-field `tsvector`, `pg_trgm` fuzzy fallback) — **always on**.
+- **Semantic** (pgvector, HNSW) + **AI-assisted** (intent parse, RAG) — **optional**.
+- **Hybrid fusion (RRF)** merges lexical+semantic; boosts recency/sender/unread.
+- **Keyset pagination** replaces offset; score-ordered top-N for ranked queries.
+- `SearchIntentParser` (evolves `GmailQueryParser`): operators → rules-NL → optional LLM.
+
+## AI architecture
+- Provider interfaces + `IInboxAi` facade + **model router** (task→model via config) +
+  **prompt templates** (versioned files) + **VRAM guard** (serialises heavy vision jobs
+  behind unloading the 7B; embeddings stay hot). See [06](06-ai-strategy.md).
+- **Streaming** via SSE for summaries/replies/RAG answers (perceived speed).
+- **Bounded**: per-call timeout, `CancellationToken`, Polly fallback to Null path.
+
+## Plugin / modularity architecture
+- **Analyzer pipeline:** email enrichment is a set of `IEmailAnalyzer` plugins
+  (classifier, entity-extractor, summariser, phishing, dedup). Each declares required
+  capabilities (`Chat`/`Embeddings`/none) and is **skipped gracefully** if unavailable.
+  New AI features = new analyzers; no core changes.
+- **Search providers** implement a common `ISearchLayer` so semantic/AI layers plug in.
+- **AI providers** already pluggable (`IAiProvider`); adding a provider = one class + config.
+- This is the "modular AI" requirement realised structurally.
+
+## Caching
+- **Embeddings & summaries**: persisted in Postgres (compute once, invalidate on new msg).
+- **Search suggestions / recent**: cached per-user (memory + DB); debounced live search.
+- **Query results**: short-TTL cache for identical repeated queries; ETag on read endpoints.
+- **Model warmth**: Ollama `keep_alive` keeps the 7B hot between calls.
+
+## Indexing
+| Index | Column | Purpose |
+|-------|--------|---------|
+| GIN | `SearchVector` (weighted A/B/C/D) | Full-text |
+| GIN `pg_trgm` | sender addr/name, subject | Fuzzy / substring (fixes non-sargable `.Contains`) |
+| HNSW | `Email.Embedding` (pgvector) | Semantic k-NN |
+| btree | `(UserId, SentAtUtc)`, `(UserId, ThreadId)` | Keyset pagination, threading |
+- Built incrementally on sync; **EmbeddingBackfillWorker** batch-fills history (VRAM-aware, low priority).
+
+## Background jobs
+- Keep the existing `IHostedService` worker pattern; add a **`Channel<T>` in-process queue**
+  with a bounded concurrency worker for AI enrichment (summaries/entities/embeddings on new
+  mail). Idempotent, resumable, backpressured. (Upgrade path: durable queue if multi-node.)
+- Jobs: `GmailSyncWorker` (exists), `DigestWorker` (exists), `EmbeddingBackfillWorker`,
+  `AiEnrichmentWorker`, `IndexMaintenanceWorker`.
+
+## Database schema improvements
+- **Widen** `Email.SearchVector` (subject/sender/filename/labels, weighted).
+- **Add**: `Email.Embedding vector(768)`; tables `EmailEntities`, `ThreadSummary`,
+  `SavedSearch`, `SearchHistory`, `AiJob` (status/retry), `SenderImportance` (materialised).
+- **Fix debt**: resolve the `Email↔EmailLabel` global-query-filter warning (optional nav or
+  matching filters); make FTS **language-aware** (detect → per-language config) instead of hardcoded English.
+- **Scale**: consider per-`UserId` list partitioning of `Emails` if single users exceed ~1M rows.
+
+## Scalability
+- Single-user/self-host is the primary shape → vertical scaling + good indexes is enough.
+- Connection pooling (Npgsql), keyset pagination, top-N-by-score, batched embeddings.
+- **Ollama is the throughput bottleneck** → serialise/queue AI, cache aggressively, prefer
+  the small/embedding models for high-volume paths.
+- Multi-tenant future: read replicas, per-user partitioning, durable job queue.
+
+## Deployment
+- Extend the Compose stack (from the CI/CD we built): add an **optional `ollama` service**
+  (profile `ai`) and a **pgvector-enabled Postgres image**. AI-off deployments omit the
+  profile entirely. Dev→staging auto-deploy already proven; prod is tag-gated.
+
+## Offline support
+- Data is **already local** (Postgres on the user's machine) — the product is local-first
+  by nature. SPA: service-worker cache for shell + last-viewed mail (read offline); queue
+  mutations (label/cleanup) to replay on reconnect. AI is inherently offline (Ollama local).
+
+## Security model
+- Retain: OAuth **read-only** scope, encrypted refresh tokens (Data Protection), **IDOR
+  global query filters**, SSRF egress guard, non-root containers, confirmed+previewed
+  destructive actions.
+- **New AI concerns:**
+  - **Prompt injection** — email content is untrusted input to the LLM. Treat model output
+    as **advisory only**; never let it trigger actions directly; sanitise/format; the LLM
+    can *suggest* but a human/rule confirms (aligns with "AI never acts").
+  - **Attachment/vision** — sandbox parsing; size/type limits; never execute.
+  - **Local-only by default** — cloud provider is explicit opt-in with per-feature consent
+    and egress logging; the SSRF guard already constrains outbound calls.
+  - Secrets stay in `deploy/.env` / Actions secrets (never in git) — as established.

docs/discovery/09-roadmap.md

@@ -0,0 +1,93 @@
+# 09 — Product Roadmap (Phase 8)
+
+Everything from discovery, split into releases and **prioritised by user value vs
+engineering effort**. The guiding sequence: **ship the redesign + dramatically better
+(deterministic) search first**, then layer **optional local AI** in value order, each tier
+reusing infrastructure the previous one built.
+
+> **Update — multi-provider platform epic.** The
+> [multi-provider + admin/settings design](multi-provider/README.md) is a **v1.x platform
+> epic** (its own 6-phase plan) that the search/AI features build on. Sequencing note: the
+> **provider abstraction, unified email model, settings, and feature-flag engine land early
+> in v1.x** (they underpin multi-user + AI gating); the AI features from this roadmap then
+> plug into that flag system. Both plans share the same flag-gated, ship-dark discipline.
+
+## Prioritisation framework
+Score each item **Value (1–5) × (6 − Effort 1–5)**, then sequence so that (a) high-value/
+low-effort ships first, (b) risky/expensive AI comes only after its infrastructure exists,
+and (c) **nothing in an early release depends on AI being enabled.**
+
+```
+ Value ▲   ★ MVP first        ● do next        ○ later
+   5 │ ★ranking ★redesign     ●NL search       ○ask-inbox
+   4 │ ★fuzzy ★chips ★summary ●semantic ●brief  ○knowledge graph
+   3 │ ★perf/indexes          ●people ●extract  ○collaboration
+   2 │                        ●categorise-tail  ○plugins ○mobile
+   1 │                                          ○multi-provider
+     └───────────────────────────────────────────────────►
+        low effort ──────────────────────────► high effort
+```
+
+---
+
+## MVP → **v1.0.0** — "The redesign + world-class deterministic search"
+Rationale: the single biggest perceived-quality jump, built mostly on **deterministic**
+tech (low risk). AI appears only as a few optional, reversible wins behind a toggle.
+- **UX v2 shell + design system + both themes** (dark-first, green ramp) — the "world-class" feel.
+- **Search core:** relevance ranking (RRF/`ts_rank`), multi-field weighted FTS, `pg_trgm`
+  fuzzy, **filter chips**, saved/recent/suggested, **"why matched"** (lexical), keyset pagination.
+- **Perf/debt:** `pg_trgm` + FTS indexes, virtualised list, code-splitting, fix EF
+  query-filter warning, language-aware FTS.
+- **AI foundation:** extended abstraction (`IAiProvider`+`IEmbeddingProvider`+facade+router),
+  Null + Ollama wired, **VRAM guard**, prompt templates.
+- **First AI wins (optional):** thread summary · follow-up detection (heuristic + AI confirm)
+  · reply suggestions.
+- **Why now:** delivers the flagship promise (fast, approachable, ranked, explainable search
+  + a premium UI) even with AI off.
+
+## v1.1 — "Assisted search & productivity"
+Rationale: build on MVP's embedding groundwork; assist without heavy compute.
+- **Natural-language search** parse (rules-first + optional LLM), shown as editable chips.
+- **People search** · **attachment (filename) search** · **search-driven bulk actions**.
+- **Inbox assistant / daily brief** · **task/calendar/reminder extraction**.
+- **Categorisation long-tail** (embedding zero-shot + confidence) · **priority prediction v1**.
+- **Why now:** the "assisted, not syntactic" search vision, plus the highest-value AI
+  productivity features — all still light on VRAM.
+
+## v1.2 — "Semantic tier"
+Rationale: introduces `pgvector` + embedding backfill; unlocks meaning-based features.
+- **Semantic search** (pgvector HNSW, hybrid RRF) · **related / find-similar**.
+- **Near-duplicate detection** · **thread summaries surfaced in results** · **conversation insights**.
+- **Relationship mapping (basics)**.
+- **Why now:** semantic recall is a headline differentiator but needs the embedding
+  infrastructure and backfill worker to be mature and VRAM-safe.
+
+## v2.0 — "Ambitious AI"
+Rationale: flagship, compute-heavy features that need a mature semantic + vision stack.
+- **Conversational "ask your inbox"** (RAG + **citations**).
+- **Entity search & facets** (amounts, orgs, dates) · **attachment OCR/content understanding** (vision, on-demand).
+- **Phishing reasoning** (LLM over flagged mail, async, explained) · **knowledge graph**.
+- **Why now:** highest wow-factor and the strongest "local private AI" story, but only
+  worthwhile once retrieval, extraction, and vision infra are proven.
+
+## v3.0 — "Platform"
+Rationale: expand beyond the individual power-user once the core is best-in-class.
+- **Collaboration / shared inbox** (assign, comment) · **automation / rules engine**
+  (a `docs/specs/feature-rules-engine.md` already exists — fold it in).
+- **Multi-account & additional providers** (IMAP/Outlook) · **plugin ecosystem** (analyzer
+  API) · **mobile app** · **multi-window / desktop shell** (Tauri) · **multi-provider AI**.
+- **Why now:** platform bets that only pay off on top of a beloved core product.
+
+---
+
+## Sequencing principles (explained)
+1. **Deterministic value before AI.** MVP's biggest wins (ranking, chips, redesign) need no
+   AI — they de-risk the release and prove the product before compute-heavy features.
+2. **Infrastructure amortised.** Embeddings introduced once (v1.1 foundations → v1.2 usage)
+   power search, dedup, related, categorisation, and RAG — spread the cost.
+3. **Value-first within a release.** Inside each version, highest value/effort ships first so
+   partial delivery is still shippable.
+4. **AI is always additive.** Every release is complete and excellent with AI disabled —
+   protecting the "works without AI" mandate and the mainstream audience.
+5. **Platform last.** Collaboration/plugins/mobile are large and only worthwhile once the
+   individual experience is genuinely best-in-class.

docs/discovery/10-git-plan.md

@@ -0,0 +1,60 @@
+# 10 — Git Implementation Plan (Phase 10)
+
+Turns the [roadmap](09-roadmap.md) into executable Git work, on the workflow already in
+place ([../WORKFLOW.md](../WORKFLOW.md)): trunked `develop`/`main`, Conventional Commits,
+PR-gated CI/Security, auto-deploy to staging, tag-gated production.
+
+## Structure: Epics → Features → Tasks
+- **Epic** = a roadmap theme → a milestone + a long-lived integration effort.
+- **Feature** = one shippable slice → one `feature/*` branch → one PR into `develop`.
+- **Task** = one atomic commit (conventional) within a feature branch.
+
+## Epics (mapped to releases)
+| Milestone | Epic | Example feature branches |
+|-----------|------|--------------------------|
+| **v1.0.0** | `epic/design-system` | `feature/design-tokens` · `feature/app-shell` · `feature/themes-dark-light` |
+| **v1.0.0** | `epic/search-core` | `feature/search-ranking` · `feature/fts-multifield` · `feature/search-fuzzy-trgm` · `feature/search-chips` · `feature/saved-recent-searches` · `feature/search-why-matched` · `feature/keyset-pagination` |
+| **v1.0.0** | `epic/perf-and-debt` | `feature/trgm-indexes` · `feature/virtualised-list` · `feature/code-splitting` · `fix/ef-queryfilter-warning` · `feature/language-aware-fts` |
+| **v1.0.0** | `epic/ai-foundation` | `feature/ai-abstraction` · `feature/embedding-provider` · `feature/ai-vram-guard` · `feature/ai-thread-summary` · `feature/ai-followup-detect` · `feature/ai-reply-suggest` |
+| **v1.1.0** | `epic/assisted-search` | `feature/nl-search-parse` · `feature/people-search` · `feature/attachment-filename-search` · `feature/search-bulk-actions` |
+| **v1.1.0** | `epic/ai-productivity` | `feature/inbox-brief` · `feature/task-calendar-extract` · `feature/categorise-tail` · `feature/priority-v1` |
+| **v1.2.0** | `epic/semantic` | `feature/pgvector-schema` · `feature/embedding-backfill` · `feature/semantic-search` · `feature/find-similar` · `feature/near-dup` · `feature/thread-insights` |
+| **v2.0.0** | `epic/ambitious-ai` | `feature/ask-your-inbox-rag` · `feature/entity-facets` · `feature/attachment-ocr` · `feature/phishing-reasoning` · `feature/knowledge-graph` |
+| **v3.0.0** | `epic/platform` | `feature/rules-engine` · `feature/collaboration` · `feature/multi-account` · `feature/plugin-api` · `feature/mobile` |
+
+## Branch naming
+- `feature/<kebab-scope>` · `fix/<kebab>` · `hotfix/<kebab>` (off `main`) · optional
+  `release/x.y.0` for stabilisation. Epics tracked via milestone/label, not a long branch
+  (avoid merge hell); features integrate continuously into `develop`.
+
+## Commit strategy
+- **Conventional Commits** (already used): `type(scope): summary`. Types drive SemVer:
+  `feat`→minor, `fix`→patch, `feat!`/`BREAKING CHANGE`→major.
+- One logical change per commit; **docs updated in the same commit/PR** as the behaviour
+  they describe (enforced by review checklist — see below).
+
+## PR strategy
+- `feature/* → develop`, **squash-merge**; `develop → main`, **merge commit** (release boundary).
+- **Required checks** (already enforced): `CI/backend`, `CI/frontend`, `Security/secrets`,
+  `Security/dependencies`. Merge to `develop` auto-deploys **staging**.
+- **PR checklist:** tests added (unit + integration for API changes) · docs updated ·
+  AI features have a **Null/AI-off path** · no secret committed · perf-sensitive paths have
+  an index/plan note.
+
+## Release milestones (tags on `main`)
+| Tag | Contents | Gate |
+|-----|----------|------|
+| `v1.0.0` | Redesign + deterministic search + AI foundation & first wins | Redesign parity + search benchmarks green |
+| `v1.1.0` | Assisted search + AI productivity | NL parse accuracy + extraction quality bar |
+| `v1.2.0` | Semantic tier | Backfill complete + hybrid-rank quality bar |
+| `v2.0.0` | Ambitious AI (RAG, vision, graph) | RAG citation accuracy + VRAM stability |
+| `v3.0.0` | Platform (collab, rules, plugins, mobile) | — |
+- Cutting a tag = the production-promotion action (see [../WORKFLOW.md](../WORKFLOW.md) §5–6);
+  `deploy-prod.yml` fires on `v*` once the Linux server + its runner exist.
+- Interim work ships as `0.x`/pre-release increments on `develop`; `v1.0.0` is the first
+  "world-class" cut.
+
+## Docs-alongside-code (hard rule)
+Every feature PR updates the relevant doc: `docs/discovery/*` decisions graduate into
+`docs/` living docs (architecture, search, AI, API reference) as they're implemented, and
+`CHANGELOG.md` gains an entry. Discovery docs are the *source*; implementation keeps them true.

docs/discovery/11-risks-and-future.md

@@ -0,0 +1,37 @@
+# 11 — Risk Assessment & Future Opportunities
+
+## Risk assessment
+Likelihood (L) / Impact (I): H/M/L.
+
+| # | Risk | L | I | Mitigation |
+|---|------|---|---|------------|
+| R1 | **Full UI redesign destabilises a working app** | M | H | Strangler migration behind `ui.v2` flag, screen-by-screen; API untouched; ship via the proven `develop→staging` pipeline; keep old screens until parity verified |
+| R2 | **VRAM (10 GB) can't hold desired models concurrently** | M | M | 7–8B sweet-spot (not 14B); embeddings hot + LLM warm + vision on-demand; VRAM guard serialises heavy jobs; small models for high-volume paths |
+| R3 | **Local AI quality/latency disappoints** | M | M | Traditional-first (AI only where it clearly wins); AI advisory + optional; stream responses; cache; model routing lets us swap models without code change |
+| R4 | **Prompt injection via email content** | M | H | Treat all model output as advisory; **AI never triggers actions**; human/rule confirms; sanitise; SSRF/egress guards; local-only by default |
+| R5 | **Gmail API quota / sync scale at 100k+ mailboxes** | M | M | Batching + Polly backoff (exists); incremental sync; background enrichment queue with backpressure; keyset pagination |
+| R6 | **Semantic infra (pgvector/embeddings) ops complexity** | M | M | Introduce once (v1.2), backfill worker VRAM-aware + resumable; HNSW tuning; feature hides if unavailable |
+| R7 | **Search relevance regressions vs today** | L | M | Lexical hits never lose to fuzzy noise (weighting); benchmark suite as a release gate; keep date-sort as a user option |
+| R8 | **Scope creep — trying to beat everyone at once** | H | M | Roadmap value/effort discipline; MVP is deterministic + small AI; platform features deferred to v3 |
+| R9 | **Solo-dev bandwidth / single-machine staging** | H | M | Small shippable features; CI/CD automation already reduces toil; staging owned by automation (don't hand-run it — see memory note) |
+| R10 | **Mainstream-vs-power tension dilutes the UX** | M | M | Progressive disclosure: simple default, power revealed on demand; pointer-first with keyboard as accelerator |
+| R11 | **Privacy promise broken by a cloud provider option** | L | H | Local default; cloud is explicit per-feature opt-in with egress logging + consent; never silent |
+| R12 | **pgvector image / Ollama container adds deploy friction** | L | L | Optional Compose profiles (`ai`); AI-off deployments omit them entirely |
+
+## Future opportunities (beyond v3.0)
+- **Additional mail backends** — IMAP/JMAP, Outlook/Graph — become a true multi-provider client.
+- **On-device personalisation** — light fine-tuning / user-preference adapters for priority & tone.
+- **Calendar & tasks integration** — close the loop from extraction to action.
+- **Voice** — dictate replies, "ask your inbox" by voice (local Whisper).
+- **Plugin marketplace** — third-party analyzers/widgets on the analyzer + command-palette APIs.
+- **Team knowledge base** — shared, permissioned knowledge graph across a team inbox.
+- **Native desktop shell** (Tauri) for OS integration, global hotkey, tray, true multi-window.
+- **Smart compose surfaces** — templates that learn, snippet library, per-recipient tone memory.
+- **Local model upgrades** — swap in newer/quantised models as they ship (router makes it a config change).
+
+## Recommendation
+The foundation is strong and the wedge is real. **Proceed with v1.0.0 (redesign +
+deterministic search + AI foundation)** — it's high-value, low-risk, and independent of AI
+being enabled — then layer local AI in value order. The biggest watch-items are **R1
+(migration discipline)** and **R8 (scope)**; both are controlled by the strangler approach
+and the value/effort-sequenced roadmap.

docs/discovery/EXECUTIVE-SUMMARY.md

@@ -0,0 +1,70 @@
+# Executive Summary — InboxIntel Discovery Blueprint
+
+**Ambition:** become one of the best **email search and management** experiences available
+— fast, scalable, secure, intuitive — with **AI used only where it clearly beats
+traditional approaches**, running **locally and privately**, and **never as a hard
+dependency**.
+
+## The opportunity (the wedge)
+The market has split in two, and both halves leave a gap:
+- **AI, but cloud** (Gmail/Gemini, Outlook/Copilot, Shortwave, Spark) — useful, but your
+  mail is processed off-device.
+- **Private, but little/no AI** (Proton, Thunderbird, Fastmail).
+
+**No mainstream product owns "genuinely useful AI that runs on your own machine."**
+InboxIntel can — and it already has the seams for it. Combined with three more openings —
+**approachable power-search as the home screen**, **explainability** ("why this matched /
+was categorised"), and **fast *and* friendly** (Superhuman is fast-but-intimidating; casual
+apps are friendly-but-slow) — this is a defensible, differentiated position.
+
+> **Positioning:** *Make search the fastest way to think about your inbox, with AI that runs
+> on your own machine and always explains itself. Powerful for pros, simple for anyone.*
+
+## Current state — verdict: extend, don't rewrite
+A clean, secure, well-tested .NET 8 / React foundation with **real Postgres full-text
+search** and **an AI provider abstraction already in place** (`NullAiProvider` /
+`OllamaProvider` / `OpenAiProvider`). The gaps are exactly where the product wants to win:
+relevance-ranked multi-mode search, a richer AI contract (embeddings/extraction),
+conversation intelligence, and a modern approachable UX. **None require a rewrite.**
+
+## Design direction (from the interview)
+Notion/Arc-**professional**, "anyone can pick it up": **pointer-first & discoverable**
+(palette/shortcuts as accelerators), **balanced density**, **dark-first** (genuine light
+too), **subtle motion**, **green `#3ba31f`** accent, clean rounded line icons, **fully
+responsive** desktop→mobile. The pivotal decision — *discoverable-first over keyboard-first*
+— reframes search from "a syntax you learn" into "an inviting, assisted experience."
+
+## Search & AI in a nutshell
+- **Search:** a 4-layer engine — Structured → **Lexical+Ranking** (always on) → Semantic
+  (pgvector) → AI-assisted (NL / ask-your-inbox / explanations) — with **hybrid RRF ranking**
+  replacing today's date-only sort. Degrades gracefully with AI off.
+- **AI:** extend the abstraction to embeddings + structured output behind a task facade with
+  **config-driven model routing**; on the **RTX 3080 (10 GB)**, **Qwen2.5-7B** (warm) +
+  **nomic-embed-text** (hot) do most jobs, vision on-demand. Traditional-first everywhere;
+  every AI feature has a non-AI fallback.
+
+## Roadmap shape
+**v1.0** redesign + world-class *deterministic* search + AI foundation → **v1.1** assisted
+search & productivity → **v1.2** semantic tier → **v2.0** ambitious AI (RAG, vision, graph)
+→ **v3.0** platform (collaboration, rules, plugins, mobile). Sequenced so **deterministic
+value ships before AI**, infrastructure is amortised, and **every release is complete with
+AI disabled.**
+
+## Key risks
+Migration of a full redesign (**mitigated by strangler + `ui.v2` flag**), VRAM limits
+(**7–8B sweet spot + load policy**), prompt injection (**AI advisory-only, never acts**), and
+scope creep (**value/effort-sequenced roadmap**). See [11](11-risks-and-future.md).
+
+## Recommendation
+**Proceed to v1.0.0** — the redesign, ranked/assisted search, and AI foundation. It's the
+biggest quality jump, it's low-risk and deterministic, and it stands entirely on its own
+without AI. Then layer local, private, explainable AI in value order.
+
+---
+### Read the full blueprint
+[01 Architecture](01-architecture-review.md) · [02 Competitors](02-competitor-analysis.md) ·
+[03 Users](03-user-research.md) · [04 UX + Design System](04-ux-redesign-and-design-system.md) ·
+[05 Search](05-search-redesign.md) · [06 AI Strategy](06-ai-strategy.md) ·
+[07 AI Features](07-ai-feature-catalogue.md) · [08 Architecture](08-technical-architecture.md) ·
+[09 Roadmap](09-roadmap.md) · [10 Git Plan](10-git-plan.md) ·
+[11 Risks & Future](11-risks-and-future.md) · [Design Brief](00-design-brief.md)

docs/discovery/README.md

@@ -0,0 +1,48 @@
+# InboxIntel — Discovery & Blueprint
+
+This folder is the **product discovery output**: a world-class specification that will
+become the blueprint for the next generation of InboxIntel. It is a **planning
+artifact** — no application code is changed during discovery.
+
+> Vision: become one of the best **email search and management** experiences available —
+> extremely fast, scalable, secure, and intuitive. AI is introduced **only where it
+> clearly beats traditional approaches**, and never as a hard dependency.
+
+## Non-negotiable constraints (carried into every phase)
+- **AI is optional & modular.** The app must work fully with AI disabled. All AI sits
+  behind an `IAIProvider` abstraction (already seeded: `NullAiProvider` / `OllamaProvider`
+  / `OpenAiProvider`) so providers swap without touching application logic.
+- **Local-first AI.** Primary target: Ollama on an NVIDIA RTX 3080 (10 GB).
+- **Privacy.** Read-only Gmail scope; AI advisory-only; no destructive AI actions.
+- **Speed & security are features**, not afterthoughts.
+
+## Document index
+> **Start here:** [Executive Summary](EXECUTIVE-SUMMARY.md)
+
+| # | Document | Phase | Status |
+|---|----------|-------|--------|
+| — | [Executive Summary](EXECUTIVE-SUMMARY.md) | all | ✅ draft |
+| 00 | [Design Brief](00-design-brief.md) | 4A (interview) | ✅ locked |
+| 01 | [Architecture Review](01-architecture-review.md) | 1 | ✅ draft |
+| 02 | [Competitor Analysis](02-competitor-analysis.md) | 2 | ✅ draft |
+| 03 | [User Research & Journey Maps](03-user-research.md) | 3 | ✅ draft |
+| 04 | [UX/UI Redesign + Design System](04-ux-redesign-and-design-system.md) | 4A | ✅ draft |
+| 05 | [Search Redesign](05-search-redesign.md) | 4B | ✅ draft |
+| 06 | [AI Strategy & Model Recommendations](06-ai-strategy.md) | 5 | ✅ draft |
+| 07 | [AI Feature Catalogue](07-ai-feature-catalogue.md) | 6 | ✅ draft |
+| 08 | [Technical Architecture](08-technical-architecture.md) | 7 | ✅ draft |
+| 09 | [Product Roadmap (MVP→v3)](09-roadmap.md) | 8 | ✅ draft |
+| 10 | [Git Implementation Plan](10-git-plan.md) | 10 | ✅ draft |
+| 11 | [Risk Assessment & Future Opportunities](11-risks-and-future.md) | cross-cutting | ✅ draft |
+
+## Extension designs
+- [**Multi-Provider Email Platform + Admin/Settings**](multi-provider/README.md) — evolves
+  InboxIntel into a small-team, multi-provider (Gmail/Outlook/IMAP) platform with OAuth-as-
+  login, settings, feature flags, and an admin panel. Reshapes the core assumptions above
+  (see the "Update" callouts in [08](08-technical-architecture.md) and [09](09-roadmap.md)).
+
+## Conventions
+- Every recommendation is a durable markdown doc suitable for long-term maintenance.
+- Each feature is specified with: **problem solved · why AI (or why not) · complexity ·
+  estimated effort · user value · perf & privacy considerations**.
+- Nothing here is implemented until explicitly approved and scheduled via the Git plan.

docs/discovery/multi-provider/01-provider-abstraction.md

@@ -0,0 +1,80 @@
+# 01 — Provider Abstraction (Part 1)
+
+A unified layer so Gmail, Outlook/Graph, and future IMAP look identical to the rest of the
+app. **No provider-specific logic in Domain**; specifics live in Infrastructure adapters.
+
+## Layering
+```
+Domain            : Account, EmailMessage, EmailThread, Label (provider-agnostic)
+Application       : IEmailProvider (contract) · ISyncOrchestrator · DTOs
+Infrastructure    : GmailProvider · OutlookProvider · ImapProvider  (adapters)
+                    ProviderFactory (ProviderType → adapter) · ITokenStore
+```
+
+## The contract
+```csharp
+public enum ProviderType { Google, Microsoft, Imap }
+
+public interface IEmailProvider {
+    ProviderType Type { get; }
+    ProviderCapabilities Capabilities { get; }          // read, modifyFlags, folders, delta, send?
+
+    // Auth (details in 02-auth-and-signin.md)
+    Task<OAuthResult> ExchangeCodeAsync(string code, CancellationToken ct);
+    Task<TokenSet>    RefreshAsync(TokenSet current, CancellationToken ct);
+    Task<ProviderIdentity> GetIdentityAsync(TokenSet tokens, CancellationToken ct); // sub + email
+
+    // Sync (pull-based, incremental)
+    Task<SyncPage> SyncAsync(SyncCursor cursor, TokenSet tokens, CancellationToken ct);
+    Task<RawMessage> FetchMessageAsync(string providerMessageId, TokenSet tokens, CancellationToken ct);
+
+    // Mutations (only if Capabilities allow; mirrors current gmail.modify scope)
+    Task ApplyFlagAsync(string providerMessageId, MailFlagChange change, TokenSet tokens, CancellationToken ct);
+}
+```
+- `SyncPage` = `{ IReadOnlyList<RawMessage> upserts, IReadOnlyList<string> deletes, SyncCursor next, bool hasMore }`.
+- `RawMessage` is the **provider-shaped** payload; a **normaliser** maps it to the domain
+  `EmailMessage`. The rest of the app never sees `RawMessage`.
+- Capabilities let the UI/engine **degrade gracefully** (e.g., an IMAP server without
+  CONDSTORE falls back to full-scan sync; no `send` today for any provider).
+
+## Provider implementations
+| Provider | API | Incremental cursor | Threading | Folders/Labels | Notes |
+|----------|-----|--------------------|-----------|----------------|-------|
+| **Gmail** | Gmail REST | `historyId` (History API) | `threadId` | labels | Reuses existing client; read + modify (no send), as today |
+| **Outlook/365** | Microsoft Graph | **delta query** `@odata.deltaLink` | `conversationId` | mailFolders | OAuth via Microsoft identity platform |
+| **IMAP** | IMAP4rev1 | `UIDVALIDITY`+`UIDNEXT`, `HIGHESTMODSEQ` (CONDSTORE/QRESYNC) | heuristic (References/In-Reply-To) | folders | Fallback = periodic UID scan if no CONDSTORE; MailKit already a dependency |
+
+## Normalisation (the unified model)
+Each adapter maps provider fields → domain via a `IMessageNormaliser`:
+| Domain field | Gmail | Graph | IMAP |
+|--------------|-------|-------|------|
+| `ProviderMessageId` | message id | message id | `UIDVALIDITY:UID` |
+| `ProviderThreadId` | threadId | conversationId | derived (References) |
+| flags (unread/star/important/trashed/inbox) | labelIds | isRead/flag/folder | `\Seen \Flagged`, folder |
+| labels/folders | labels | mailFolders | folders |
+| sent/received, from, subject, snippet, body, attachments, size | headers/parts | message resource | RFC822 parse (MailKit) |
+- **Threads are per-account** (each provider defines its own). *Cross-provider* thread
+  linking is a later semantic/AI feature (see blueprint [07](../07-ai-feature-catalogue.md)),
+  not part of core normalisation.
+
+## Sync engine
+- **`ISyncOrchestrator`** replaces the Gmail-specific worker: for each active `Account`, it
+  loads the `SyncCursor`, calls `provider.SyncAsync`, **upserts** normalised messages
+  (idempotent on `(AccountId, ProviderMessageId)`), applies deletes, and **persists the next
+  cursor** atomically.
+- Runs as the existing hosted-worker pattern (`AccountSyncWorker`), one logical job per
+  account, bounded concurrency, Polly backoff, resumable.
+- **Token refresh:** `SyncAsync`/mutations get a valid `TokenSet` from `ITokenStore`, which
+  refreshes on expiry/401 and **re-encrypts** at rest; a failed refresh flips the account to
+  `reauth_needed` (surfaced in UI, see [07](07-ux-flows.md)) — never crashes sync.
+- **New mail** triggers AI enrichment + embedding jobs (blueprint [08](../08-technical-architecture.md)).
+
+## Search across providers
+Because all providers normalise into **one `email_messages` store scoped by `UserId`**,
+search (structured + FTS + semantic) **already spans every account a user has connected** —
+no per-provider search code. An optional `AccountId` facet lets users scope to one mailbox.
+
+## Extensibility
+Adding a provider = one `IEmailProvider` adapter + one normaliser + register in
+`ProviderFactory` + a feature flag to enable it. **Zero changes to Domain, search, or AI.**

docs/discovery/multi-provider/02-auth-and-signin.md

@@ -0,0 +1,70 @@
+# 02 — Auth & Sign-in (Part 2)
+
+**OAuth is the login.** No passwords. A user's identity is the set of provider accounts
+linked to them; any one can authenticate the session. Identity key is **`(provider, sub)`**
+— never email (emails change; `sub` is stable, and the same email can exist on Google *and*
+Microsoft as distinct accounts).
+
+## Provider-selection sign-in (first-time)
+```
+[ Choose how to sign in ]
+   ▸ Continue with Google        ▸ Continue with Microsoft     ( ▸ IMAP — future )
+```
+1. User picks a provider → redirect to provider OAuth (**PKCE**, `state`, `nonce`).
+2. Callback → exchange code → `GetIdentityAsync` returns `(provider, sub, email, name)`.
+3. **Resolve:** look up `accounts (provider, sub)`.
+   - **No match →** first-time. Create `user` (**first user ever = Admin**, otherwise `Member`
+     if `system_settings.registration_open`, else reject) + `account` (`is_login_identity=true`)
+     + encrypted `provider_tokens`. Start session.
+   - **Match →** existing user. Refresh tokens, start session.
+4. Kick off the account's initial sync.
+
+## Adding another account later (linking) — the security-critical flow
+The user is **already authenticated**. "Add account" → provider OAuth in **link mode**:
+- Callback identity `(provider, sub)`:
+  - **Unlinked →** attach a new `account` (mailbox) to the **current** user. ✅
+  - **Already linked to *this* user →** no-op / "already connected." 
+  - **Already linked to *another* user →** **blocked** by the unique `(provider, provider_account_id)`
+    constraint + explicit check → error "This mailbox is connected to a different InboxIntel
+    user." **This is the anti-hijack guarantee** — you can only link an identity you can
+    authenticate *and* that no one else owns.
+- A single user can hold **N accounts** across Google/Microsoft/IMAP; each can also serve as a
+  login identity (any of them signs you into the same user).
+
+## Switching
+- **Switch mailbox (same user):** an **account switcher** changes the active mailbox context
+  (or "All accounts" unified view). No re-auth — it's all one user. Search can scope to one
+  account or span all.
+- **Switch user (different person):** full sign-out → sign-in. Optional "fast switch" could
+  hold multiple sessions, but for a small team, explicit re-auth is simplest and safest.
+
+## Sessions
+- **Opaque server-side session** (`sessions` table) referenced by an **HttpOnly · Secure ·
+  SameSite=Lax** cookie. **Provider tokens are never exposed to the browser.**
+- Rotate session id on login (anti-fixation); **idle (e.g., 7d) + absolute (e.g., 30d)**
+  expiry; revoke on logout; **"sign out everywhere"** and **admin revoke** delete session rows.
+- CSRF: SameSite + anti-CSRF token on state-changing requests.
+
+## Token lifecycle
+- Stored **encrypted at rest** (Data Protection); decrypted only in-memory for API calls.
+- **Refresh** on expiry/401 via `ITokenStore` → re-encrypt + persist; failure flips account to
+  **`ReauthNeeded`** (banner + "Reconnect" CTA, see [07](07-ux-flows.md)) — sync/AI for that
+  account pause, the rest of the app is unaffected.
+
+## Provider OAuth specifics
+| | Google | Microsoft (Graph) |
+|--|--------|-------------------|
+| Endpoint | accounts.google.com | login.microsoftonline.com (`common`) |
+| Scopes | `openid email profile gmail.readonly gmail.modify` | `openid email profile offline_access Mail.Read Mail.ReadWrite` |
+| Identity | `sub` (+ verified email) | `oid`/`sub` (+ email) |
+| Refresh | refresh_token (offline) | refresh_token (`offline_access`) |
+| Redirect | `/signin/google` | `/signin/microsoft` |
+- **Least privilege:** request read/modify only (no send today — matches current posture).
+  Extra scopes are added per-feature with consent, never up-front.
+
+## Edge cases
+- **Same email, two providers** → two distinct accounts (identity is `sub`), unless the user
+  links both to one InboxIntel user.
+- **Provider disabled by admin flag** (`provider.microsoft=false`) → hide it on the picker;
+  existing accounts of that provider pause sync and show a notice.
+- **Reused browser / stale cookie** → session validated server-side each request; revoked/expired → re-auth.

docs/discovery/multi-provider/03-database-design.md

@@ -0,0 +1,79 @@
+# 03 — Database Design (Part 6)
+
+Schema for a **small-team, self-hosted, one-org** platform: multi-provider accounts per
+user, a normalised email store scaling to **millions of messages**, settings, feature
+flags, and audit. PostgreSQL + EF Core (+ pgvector for the blueprint's semantic tier).
+
+## Entity map
+```
+users ─┬─< accounts ─┬─< provider_tokens (1:1, encrypted)
+       │             ├─< email_threads ─< email_messages ─┬─< attachments
+       │             │                                    └─< message_labels >─ labels
+       │             └─ sync_state (cursor)
+       ├─< user_settings (1:1)
+       └─< audit_logs (actor)
+system_settings (singleton)      feature_flags
+```
+
+## Tables
+
+### Identity & access
+- **`users`** — the app identity (from OAuth).
+  `id (uuid pk) · primary_email (citext, unique) · display_name · avatar_url · role (enum: Admin|Member) · status (enum: Active|Suspended) · created_at · last_login_at`
+  *First-ever user is bootstrapped as **Admin** (see [05](05-admin-system.md)).*
+- **`accounts`** — a connected mailbox **and** a login identity (OAuth-is-login).
+  `id (uuid pk) · user_id (fk) · provider (enum: Google|Microsoft|Imap) · provider_account_id (text, the OAuth 'sub' — immutable) · email (citext) · display_name · is_login_identity (bool) · status (enum: Active|ReauthNeeded|Disabled) · scopes (text[]) · added_at · last_sync_at`
+  **Unique:** `(provider, provider_account_id)` → resolves an OAuth login to exactly one account→user; prevents the same mailbox linking twice.
+- **`provider_tokens`** — 1:1 with `accounts`, **encrypted at rest** (Data Protection API).
+  `account_id (pk/fk) · access_token_enc (bytea) · refresh_token_enc (bytea) · expires_at_utc · token_type · rotated_at`
+  *Never logged; see [06](06-security-model.md).*
+- **`sessions`** — server-side app sessions (opaque cookie).
+  `id · user_id · created_at · expires_at · ip · user_agent · revoked_at` (supports "sign out everywhere" + admin revoke).
+
+### Email (normalised, provider-agnostic)
+- **`email_threads`** — per account.
+  `id (uuid pk) · account_id (fk) · user_id (denorm) · provider_thread_id · subject · participants (jsonb) · message_count · last_message_at`
+  **Unique:** `(account_id, provider_thread_id)`.
+- **`email_messages`** — the big table.
+  `id (uuid pk) · account_id (fk) · user_id (denorm) · thread_id (fk) · provider_message_id · sender_id (fk) · subject · snippet · body_text · sent_at_utc · received_at_utc · size_bytes · flags (unread/starred/important/in_inbox/trashed as bits or bools) · has_attachments · category (enum) · has_list_unsubscribe · supports_one_click_unsub · search_vector (tsvector, generated, weighted) · embedding (vector(768), nullable) · created_at`
+  **Unique:** `(account_id, provider_message_id)` (idempotent upsert).
+- **`labels`** (`id · account_id · provider_label_id · name · type`) + **`message_labels`** (`message_id · label_id`, pk both).
+- **`attachments`** (`id · message_id · filename · mime · size · provider_attachment_id · content_text nullable` for future OCR/search).
+- **`senders`** / **`domains`** (existing) — kept, scoped per user (or global with per-user stats materialised in `sender_importance`).
+
+### Settings, flags, audit
+- **`user_settings`** — 1:1 with `users`.
+  `user_id (pk) · theme (enum: system|light|dark) · inbox_layout (jsonb) · notifications (jsonb) · ai_prefs (jsonb) · provider_prefs (jsonb) · updated_at`
+- **`system_settings`** — singleton (org-wide, admin-managed).
+  `id (const) · maintenance_mode (bool) · default_theme · registration_open (bool) · updated_by · updated_at` (+ arbitrary `values jsonb` for growth).
+- **`feature_flags`** — the flag system (drives AI gating).
+  `key (pk text) · enabled (bool) · scope (enum: SystemOnly|UserOverridable) · description · rollout (jsonb, e.g. per-role) · updated_by · updated_at`
+  Seeded flags: `ai.enabled`, `ai.semantic_search`, `ai.ask_inbox`, `provider.google`, `provider.microsoft`, `provider.imap`, `maintenance.readonly`.
+- **`audit_logs`** — admin + security events.
+  `id (bigserial) · actor_user_id (fk, nullable for system) · action (text) · target_type · target_id · metadata (jsonb) · ip · created_at`
+  Append-only; indexed on `(created_at)`, `(actor_user_id)`, `(action)`.
+
+## Indexing & performance
+| Index | Column(s) | For |
+|-------|-----------|-----|
+| GIN | `email_messages.search_vector` | FTS |
+| GIN `pg_trgm` | sender/subject | fuzzy (fixes non-sargable `.Contains`) |
+| HNSW | `email_messages.embedding` | semantic k-NN |
+| btree | `(user_id, sent_at_utc desc)`, `(account_id, sent_at_utc desc)`, `(thread_id)` | keyset pagination, scoping, threading |
+| unique | `(account_id, provider_message_id)`, `(provider, provider_account_id)` | idempotency, login resolution |
+
+## Scalability (millions of emails, multi-account, incremental)
+- **Multi-account** = first-class via `accounts`; every email carries `account_id` + denormalised `user_id` (so cross-account per-user search is a single indexed scan).
+- **Millions of rows:** keyset (cursor) pagination, top-N-by-score ranking, `pg_trgm`/GIN/HNSW indexes. If a single user exceeds ~1–2M messages, **list-partition `email_messages` by `account_id`** (or hash by `user_id`).
+- **Incremental sync:** per-account `sync_state` cursor (`historyId` / `deltaLink` / `uidvalidity+modseq`) → only deltas fetched; idempotent upserts on the unique key.
+- **Idempotency & resumability:** all sync writes keyed on `(account_id, provider_message_id)`; cursor advanced atomically with the batch.
+
+## RBAC in the schema
+Small-team model = a `role` column on `users` (`Admin|Member`) — no separate roles/permissions tables yet. **Extensible** to `roles`/`permissions`/`org_id` if this ever grows to multi-tenant, without reshaping the email tables.
+
+## Migration from today (see full [Migration Guide](12-migration-guide.md))
+1. Create `users` from existing OAuth identity; set first user = **Admin**.
+2. Create one `Google` **`account`** per existing user; move current Gmail tokens → `provider_tokens`.
+3. Backfill `email_messages.account_id`/`user_id`, rename/extend from the current `Email` table; widen `search_vector`; add nullable `embedding`.
+4. Add `user_settings`, `system_settings`, `feature_flags` (seed `ai.enabled` from the current `Ai:Mode`), `audit_logs`.
+5. All additive + backfill; no destructive step — safe to run behind a maintenance window.

docs/discovery/multi-provider/04-settings-and-flags.md

@@ -0,0 +1,68 @@
+# 04 — Settings & Feature Flags (Part 3)
+
+Two layers of configuration — **per-user preferences** and **org-wide system settings /
+feature flags** — with a clear precedence. Critically: **AI is governed by a feature flag
+(admin, global), not merely a user preference.**
+
+## User settings (per user)
+Stored in `user_settings`; editable by the user.
+| Setting | Values |
+|---------|--------|
+| `theme` | system · light · dark (dark-first default) |
+| `inbox_layout` | density, pane layout, default view/lane, per-account or unified |
+| `notifications` | channels, quiet hours, priority-only |
+| `ai_prefs` | per-feature opt-in (summaries, replies, semantic, ask-inbox…) — **only effective if the flag allows** |
+| `provider_prefs` | default account, sync frequency, signature per account |
+
+## System settings (admin, org-wide)
+Stored in `system_settings` (singleton); Admin-only ([05](05-admin-system.md)).
+- `maintenance_mode` (off · read-only · locked-except-admin) · `default_theme` ·
+  `registration_open` · org display name · retention defaults.
+
+## Feature flag system
+`feature_flags` rows: `key · enabled · scope · rollout · description · updated_by/at`.
+- **`scope = SystemOnly`** — a hard org-wide switch; users cannot override (e.g., `provider.microsoft`, `maintenance.readonly`).
+- **`scope = UserOverridable`** — a default that a user preference can turn *off* (never *on* beyond what the flag permits) — e.g., `ai.summaries`.
+- **`rollout`** (jsonb) — optional per-role/percentage gating (e.g., enable a beta for Admins first).
+
+### Evaluation service
+```csharp
+public interface IFeatureFlags {
+    bool IsEnabled(string key, UserContext user);   // system flag ∧ scope ∧ role rollout
+}
+public interface IAiGate {                          // the AI-specific resolver
+    bool IsAiEnabled(UserContext user);             // ai.enabled (system) ∧ user.ai_prefs.master
+    bool IsAiFeatureEnabled(string feature, UserContext user); // ∧ ai.<feature> ∧ user opt-in
+}
+```
+- Flags are **cached** with change notification (hot-reload on admin edit); every read is cheap.
+- All flag reads are **fail-closed**: unknown/errored flag ⇒ treated as **off**.
+
+## AI gating precedence (the key requirement)
+Effective AI availability is an **AND** down a chain — the system flag is the master gate:
+```
+AI feature X is available for user U  ⇔
+    feature_flags["ai.enabled"].enabled              (admin master switch — SYSTEM)
+  ∧ feature_flags["ai." + X].enabled                 (per-feature flag — SYSTEM)
+  ∧ providerCapability(X)                             (Ollama/provider actually available)
+  ∧ user.ai_prefs.master_opt_in                       (user hasn't disabled AI for themselves)
+  ∧ user.ai_prefs[X]                                  (user opted into this feature)
+```
+- **Admin turns `ai.enabled` off ⇒ AI vanishes for everyone**, regardless of any user
+  preference. This is the behaviour the brief mandates.
+- With AI on at the system level, users still choose per-feature. The **Null AI provider +
+  capability flags** (blueprint [06](../06-ai-strategy.md)) mean a disabled path **falls back
+  or hides** — never errors, never blocks core email.
+
+## Settings precedence (general)
+```
+system default  →  feature flag (may hard-disable)  →  user override (only where UserOverridable)
+```
+
+## Maintenance mode
+- `read-only`: mutations (send/cleanup/label) blocked with a banner; browsing/search stay up.
+- `locked-except-admin`: only Admins can use the app (for migrations/upgrades).
+- Enforced at the API via a middleware policy + surfaced as a global banner in the UI.
+
+## Auditing
+Flag and system-setting changes are **admin actions → audit-logged** (who/what/old→new/when).

docs/discovery/multi-provider/05-admin-system.md

@@ -0,0 +1,45 @@
+# 05 — Admin System (Part 4)
+
+An Admin-only panel to run the instance. **Admins manage the platform, not people's
+inboxes** — no admin route can read another user's mail (see [06](06-security-model.md)).
+
+## Sections
+| Section | Admin can | Notes |
+|---------|-----------|-------|
+| **Users** | List users; view role/status/last-login; **promote/demote** (Admin↔Member); **suspend/reactivate**; **revoke sessions**; remove user (with data-deletion policy) | **Never** view a user's email contents |
+| **Feature flags** | List all flags; toggle `enabled`; set scope/rollout; per-role rollout | Includes AI + provider flags |
+| **AI (global)** | Master `ai.enabled` toggle + per-feature (`ai.summaries`, `ai.semantic_search`, `ai.ask_inbox`…); see Ollama/model status | Off ⇒ AI hidden for everyone ([04](04-settings-and-flags.md)) |
+| **Providers** | Enable/disable `provider.google` / `provider.microsoft` / `provider.imap` | Disabled ⇒ hidden on login picker; existing accounts pause |
+| **System config** | Maintenance mode (off/read-only/locked); `registration_open`; org name; default theme; retention | Sensitive → step-up + audit |
+| **Monitoring** | Basic health overview (below) | Read-only |
+| **Audit log** | Search/filter admin + security events | Append-only |
+
+## Monitoring overview (basic)
+- **Sync health:** per-account last-sync time, `ReauthNeeded` count, error rate; job-queue depth.
+- **AI/Ollama:** reachable? loaded models, VRAM headroom, recent latency, failure rate.
+- **Sessions:** active session count; recent logins.
+- **System:** DB size / message count; background-job backlog; recent errors (from Serilog).
+- Deliberately **overview-only** — deep observability is a future opportunity, not v1.
+
+## Access control & bootstrap
+- Every admin route requires the **Admin policy**; sensitive mutations require **confirmation/
+  step-up** + are **rate-limited** and **audited**.
+- **Bootstrap:** the first user to sign in becomes **Admin** (one-time). Afterwards, admin is
+  granted only by an existing Admin (audited, forces target session refresh so new/removed
+  privileges take effect immediately).
+- Guardrails: an Admin cannot demote/suspend the **last remaining Admin** (lock-out prevention).
+
+## Audit logging (what's recorded)
+Actor · action · target (user/flag/setting/provider) · old→new · ip · timestamp — for **all**
+admin mutations and security events (role change, flag toggle, provider disable, maintenance
+on/off, session revoke, user suspend). Append-only `audit_logs`; visible in the Audit section;
+exportable.
+
+## API surface (Admin-scoped, all audited)
+```
+GET/PATCH  /admin/users            /admin/users/{id}/role   /admin/users/{id}/status
+GET/PATCH  /admin/flags            /admin/flags/{key}
+GET/PATCH  /admin/system-settings
+GET        /admin/monitoring       /admin/audit
+```
+All behind the Admin policy + maintenance-aware middleware.

docs/discovery/multi-provider/06-security-model.md

@@ -0,0 +1,79 @@
+# 06 — Security Model (Part 5)
+
+Builds on the existing hardening (read-only scope, encrypted tokens, IDOR global query
+filters, SSRF egress guard, non-root containers, confirmed destructive actions) and adds
+what multi-user + admin + multi-provider require.
+
+## RBAC
+- Roles: **Admin** · **Member** (small-team, one org). First user bootstraps as Admin.
+- Enforced by **policy-based authorization** at the API (ASP.NET Core policies), not in the UI.
+
+| Capability | Member | Admin |
+|------------|:------:|:-----:|
+| Read/manage **own** mail & accounts | ✅ | ✅ |
+| Own user settings | ✅ | ✅ |
+| Link/unlink **own** provider accounts | ✅ | ✅ |
+| View/manage **other users** | ❌ | ✅ |
+| Toggle **feature flags** (incl. AI global) | ❌ | ✅ |
+| Enable/disable **providers** | ❌ | ✅ |
+| **Maintenance mode**, system settings | ❌ | ✅ |
+| View **audit log** & monitoring | ❌ | ✅ |
+- **No cross-user data access, ever** — Admin manages *accounts/flags/system*, **not** other
+  users' email contents (privacy). Admin power is over the *platform*, not people's inboxes.
+
+## OAuth token storage
+- Refresh/access tokens **encrypted at rest** with the Data Protection API (AES); keys persist
+  to the mounted `/keys` volume (existing). Decrypted **only in-memory** for the moment of an
+  API call. **Never logged, never sent to the browser.**
+- 1:1 `provider_tokens` per account; rotation timestamped; a compromised/rotated token is
+  replaced atomically. Token columns are `bytea` ciphertext, not readable in DB dumps.
+
+## Session handling
+- Opaque **server-side sessions** (DB-backed) + HttpOnly/Secure/SameSite cookie; **id rotated
+  on login** (anti-fixation); idle + absolute expiry; server-side **revocation** (logout,
+  sign-out-everywhere, admin revoke, role change). CSRF via SameSite + token.
+
+## Admin access protection
+- Admin routes require the **Admin policy**; sensitive mutations (toggle AI global, disable a
+  provider, suspend a user, enter maintenance) require a **confirmation / step-up** and are
+  **rate-limited**.
+- **Every admin action is audit-logged** (`audit_logs`: actor, action, target, metadata, ip,
+  time) — append-only.
+- First-admin bootstrap is one-time; afterwards admin is grant-only by an existing Admin
+  (logged). Guard against privilege escalation: role changes are Admin-only + audited + force
+  session refresh.
+
+## API security boundaries
+- **Per-user isolation** via EF **global query filters** (extended to `account_id`/`user_id`)
+  so a query can *never* return another user's rows — the IDOR safeguard, now multi-account.
+- **Input validation** (FluentValidation) on all DTOs; **mass-assignment safe** (explicit DTOs,
+  no entity binding).
+- **Rate limiting** on auth, admin, search, and AI endpoints.
+- **SSRF egress guard** (existing) constrains all outbound calls — provider APIs, IMAP hosts,
+  Ollama, and any opt-in cloud AI — to an allowlist; user-supplied IMAP hosts are validated.
+- **Security headers** (CSP, HSTS, X-Frame-Options, etc.) via the reverse proxy/API; strict CORS.
+
+## Multi-provider & AI specifics
+- **Least-privilege scopes** per provider; extra scopes added per-feature with consent.
+- **Provider isolation:** disabling a provider flag revokes its use cleanly; per-account tokens
+  are independent (one reauth doesn't affect others).
+- **Prompt injection:** email content is untrusted → LLM output is **advisory only, never
+  triggers actions**; a human/rule confirms. AI runs **local by default**; cloud AI is explicit
+  opt-in with per-feature consent + egress logging.
+- **Attachments/vision:** sandboxed parsing, size/type limits, never executed.
+
+## Threat model (summary)
+| Threat | Mitigation |
+|--------|------------|
+| Account hijack via linking | Must authenticate as target user; unique `(provider, sub)`; linking an owned identity blocked |
+| Token theft / DB exposure | Encryption at rest; tokens never in logs/browser; rotation |
+| Privilege escalation | Admin-only role changes, audited, session refresh; policy checks server-side |
+| IDOR / cross-user leakage | Global query filters on user_id/account_id |
+| CSRF / session fixation | SameSite + token; session id rotation; server-side revoke |
+| SSRF (providers/IMAP/AI) | Egress allowlist guard; validate user-supplied hosts |
+| Prompt injection | AI advisory-only; never acts; local-first |
+| Mass admin abuse | Rate limit + step-up + full audit trail |
+
+## Non-negotiables
+Admins manage the platform, **not** users' inboxes · tokens encrypted & browser-invisible ·
+every privileged action audited · AI never required and never acts autonomously.

docs/discovery/multi-provider/07-ux-flows.md

@@ -0,0 +1,75 @@
+# 07 — UX Flows (Part 7)
+
+Applies the [Design Brief](../00-design-brief.md) (Notion/Arc-professional, dark-first,
+green accent, pointer-first, responsive). **Keep it simple** — these are utility surfaces,
+not the daily driver.
+
+## 1. Provider selection on login
+A calm, centred card — the *only* thing on screen.
+```
+            InboxIntel
+   ───────────────────────────
+   Sign in to continue
+   [  ▸  Continue with Google      ]
+   [  ▸  Continue with Microsoft   ]
+   (  IMAP — coming soon, disabled )
+   ───────────────────────────
+   Your email stays on your machine.
+```
+- Only **enabled** providers show (driven by `provider.*` flags).
+- One click → provider OAuth → back into the app. First-timer lands on an empty, friendly
+  inbox with a "syncing your mail…" state.
+
+## 2. Connected accounts page
+Reached from the account switcher or Settings → Accounts.
+```
+Accounts                                   [ + Add account ]
+┌─────────────────────────────────────────────────────────┐
+│ 🟢 Google   me@gmail.com        Synced 2m ago   ⋯        │
+│ 🟢 Microsoft me@outlook.com     Synced 5m ago   ⋯        │
+│ 🟠 Google   old@gmail.com   Reconnect needed → [Reconnect]│
+└─────────────────────────────────────────────────────────┘
+```
+- Status badges: 🟢 Active · 🟠 ReauthNeeded (with **Reconnect**) · ⚪ Disabled.
+- Per-account row menu (⋯): set as default · sync now · rename · **remove account** (confirm +
+  explains local data deletion).
+- **+ Add account** → provider picker → OAuth in link-mode → new row appears.
+- **Account switcher** (top bar): "All accounts" (unified) or pick one to scope the inbox/search.
+
+## 3. Settings page (user)
+Left sub-nav, one panel at a time — no overwhelm.
+```
+Settings
+  Appearance    ▸ Theme (System/Light/Dark) · density · layout
+  Accounts      ▸ (the page above)
+  Notifications ▸ channels · quiet hours · priority-only
+  AI            ▸ (visible only if ai.enabled; see below)
+  Privacy       ▸ data, export, clear search history
+```
+
+## 4. AI toggle visibility
+- If **`ai.enabled` (system) is OFF** → the **AI section is hidden entirely** (or shown as a
+  single disabled note: "AI features are turned off by your administrator"). No dead toggles.
+- If **ON** → a master **"Use AI features"** switch (user opt-in) + per-feature toggles
+  (Summaries · Reply suggestions · Semantic search · Ask your inbox), each reflecting its
+  `ai.<feature>` flag. Toggling off a feature instantly falls back to the non-AI path.
+- A small **status chip** ("Local · Ollama · ready") reassures it's on-device.
+
+## 5. Admin dashboard
+Only visible to Admins (nav item appears for the Admin role).
+```
+Admin
+  Overview   ▸ sync health · Ollama/VRAM · sessions · errors (cards)
+  Users      ▸ table: name · role · status · last login · [actions]
+  Flags      ▸ toggles grouped: AI · Providers · Maintenance
+  System     ▸ maintenance mode · registration · defaults
+  Audit      ▸ searchable event log
+```
+- Clean tables + toggles; destructive/sensitive actions show a **confirm dialog** (step-up).
+- Maintenance mode shows a **global banner** to all users while active.
+
+## Cross-cutting
+- **Reduced-motion & keyboard** reachability on all of the above (baseline a11y).
+- **Empty/loading/error states** per the design system (skeletons, friendly empties, calm errors).
+- Fully **responsive**: settings/admin sub-nav collapses to a top tab bar on mobile; the login
+  card is centred on all sizes.

docs/discovery/multi-provider/08-ai-feature-flags.md

@@ -0,0 +1,68 @@
+# 08 — AI Feature-Flag Integration (Part 8)
+
+How the AI layer plugs into the flag system while staying **completely separable** from core
+email logic. Extends the blueprint AI strategy ([../06](../06-ai-strategy.md)); the gate
+math lives in [04](04-settings-and-flags.md).
+
+## Principle: AI is a guest, never a host
+Core email (sync, search-lexical, cleanup, settings, admin) **never references an AI type**.
+It calls domain services; those *optionally* consult AI through a single gate + facade. Remove
+AI entirely and nothing in the core path breaks.
+
+```
+Core feature code
+   │ (never touches Ollama/IAiProvider directly)
+   ▼
+IAiGate.IsAiFeatureEnabled("summaries", user)   ──► false ─► non-AI path / hide
+   │ true
+   ▼
+IInboxAi facade (Application)  ──►  model router  ──►  IAiProvider / IEmbeddingProvider
+                                                        (Null | Ollama | future)
+```
+
+## Toggle behaviour (flag-driven)
+- Before *any* AI call, code asks `IAiGate` (which folds in `ai.enabled` + `ai.<feature>` +
+  provider capability + user opt-in — the AND-chain from [04](04-settings-and-flags.md)).
+- **Admin `ai.enabled` = off** ⇒ gate returns false everywhere ⇒ AI UI hidden, AI code paths
+  skipped. Flip on ⇒ features reappear (hot-reloaded flag cache) with **no redeploy**.
+- Per-feature flags allow shipping AI features **dark** and enabling gradually (rollout).
+
+## Fallback when AI is disabled (per feature)
+| AI feature | Fallback with AI off |
+|------------|----------------------|
+| Semantic / NL search | Lexical + structured + fuzzy search (still excellent) |
+| Thread summary | Hidden; show first snippet + metadata |
+| Reply suggestions | Hidden; normal compose |
+| Follow-up detection | Heuristic-only (sent + question + no reply in N days) |
+| Categorisation | `HeuristicClassifier` rules only |
+| Ask-your-inbox | Feature hidden |
+| Dedup | Exact-hash only (no near-dup) |
+Every fallback is **first-class**, not a broken/greyed feature — this satisfies "AI must never
+be required for core functionality."
+
+## Ollama integration layer (local models)
+- `OllamaProvider` (`IAiProvider`) + `OllamaEmbeddingProvider` (`IEmbeddingProvider`) talk to a
+  local Ollama (own container, optional Compose `ai` profile).
+- **Model router** maps logical task → model via config (`Ai:Models:{Chat,Embed,Vision}`);
+  swapping a model is a config change, not code.
+- **VRAM guard** (RTX 3080 / 10 GB): embeddings hot, 7B warm, vision on-demand
+  ([../06](../06-ai-strategy.md)).
+- **Health surfaced to admin** ([05](05-admin-system.md)): reachable? models loaded? VRAM?
+  latency? If Ollama is down, capability = false ⇒ gate falls back gracefully (no user errors).
+
+## Safe abstraction (`IAIProvider`) — separation guarantees
+1. **Interface boundary:** only Infrastructure implements providers; Application depends on
+   `IInboxAi`/`IAiGate` abstractions.
+2. **Null objects:** `NullAiProvider`/`NullEmbeddingProvider` return "unavailable" so the DI
+   graph is always valid, AI on or off.
+3. **Analyzer pipeline:** AI enrichers (`IEmailAnalyzer`) declare required capabilities and are
+   **skipped** when unavailable — adding/removing AI features never touches core sync/search.
+4. **Bounded:** every AI call has timeout + `CancellationToken` + Polly fallback to the Null
+   path; AI can never hang or crash the core.
+5. **Provider-swap:** adding a future AI provider = one class + config + (optionally) a flag —
+   no feature-code changes.
+
+## Precedence recap (single source of truth)
+The effective availability chain and admin master-switch semantics are defined once in
+[04 — Settings & Feature Flags](04-settings-and-flags.md#ai-gating-precedence-the-key-requirement);
+this document is the *architecture* of how features consume that decision.

docs/discovery/multi-provider/09-implementation-plan.md

@@ -0,0 +1,57 @@
+# 09 — Implementation Plan (Part 9)
+
+Six phases (from the brief), each **shippable behind feature flags** so `main`/staging never
+break and providers activate only when ready. Follows the established
+[../../WORKFLOW.md](../../WORKFLOW.md) pipeline (PR → checks → staging → tag→prod).
+
+## Phase 1 — Provider abstraction + Google login refactor
+- **Goal:** introduce the seam and move the *existing* Gmail behaviour behind it, with
+  OAuth-as-login and the multi-user identity foundation.
+- **Deliverables:** `IEmailProvider` + `ProviderFactory`; **`GmailProvider`** adapter wrapping
+  today's Gmail code; `users` / `accounts` / `provider_tokens` / `sessions` tables;
+  OAuth-as-login for Google; first-user→Admin bootstrap.
+- **Flags:** `provider.google` (on). **Exit:** existing Gmail users function unchanged through
+  the new abstraction; sign-in creates a `user`+`account`; all tests green.
+
+## Phase 2 — Microsoft Outlook integration
+- **Goal:** prove the abstraction with a second provider.
+- **Deliverables:** **`OutlookProvider`** (Microsoft Graph, delta query); Microsoft OAuth login
+  + link-mode; scope config; normaliser mappings.
+- **Flags:** `provider.microsoft` (**off** until verified, then rollout). **Exit:** a user can
+  link an Outlook mailbox; it syncs and searches alongside Gmail; no core changes needed.
+
+## Phase 3 — Unified email model + sync engine
+- **Goal:** formalise the normalised store and provider-agnostic sync.
+- **Deliverables:** normalised `email_messages`/`email_threads` (widened `search_vector`,
+  nullable `embedding`); **`ISyncOrchestrator`** + `AccountSyncWorker` (per-account cursors,
+  idempotent upserts, incremental); **data migration** of existing Gmail rows → default account.
+- **Flags:** none user-facing; migration behind a maintenance window. **Exit:** all providers
+  sync through one orchestrator into one store; cross-account search works.
+
+## Phase 4 — Settings system
+- **Goal:** user + system settings + the flag engine.
+- **Deliverables:** `user_settings`, `system_settings`, **`feature_flags`** + `IFeatureFlags`/
+  `IAiGate` (cached, fail-closed); settings UI; maintenance-mode middleware.
+- **Flags:** self-hosting (the engine that hosts the rest). **Exit:** users edit prefs; admins
+  can flip flags; AI gate resolves the AND-chain.
+
+## Phase 5 — Admin panel + feature flags
+- **Goal:** the Admin surface + RBAC + audit.
+- **Deliverables:** Admin API (policy-gated) + UI (Users/Flags/AI/Providers/System/Monitoring/
+  Audit); `audit_logs`; role management; basic monitoring.
+- **Flags:** admin nav shown by role. **Exit:** an Admin can manage users/flags/providers,
+  every action audited; step-up + rate-limit enforced.
+
+## Phase 6 — AI integration layer
+- **Goal:** wire optional AI behind the gate.
+- **Deliverables:** extend `IAiProvider` (+`CompleteStructuredAsync`, `IEmbeddingProvider`);
+  `IInboxAi` facade + model router + VRAM guard; analyzer pipeline; first AI features
+  (summaries, reply, follow-up-confirm) each **flag-gated + fallback**.
+- **Flags:** `ai.enabled` + `ai.<feature>` (rollout). **Exit:** AI features work when enabled,
+  **vanish/fallback** when off; core unaffected; Ollama health in admin.
+
+## Sequencing notes
+- Phases 1–3 are the platform spine; 4–5 the control plane; 6 the optional intelligence.
+- **Nothing activates on merge** — flags gate everything, so partial phases are safe on `develop`/`main`.
+- Aligns with the blueprint roadmap ([../09](../09-roadmap.md)): this multi-provider work is a
+  **v1.x platform epic** that the search/AI features then build on.

docs/discovery/multi-provider/10-git-workflow.md

@@ -0,0 +1,50 @@
+# 10 — Git Workflow (Part 11)
+
+Extends [../../WORKFLOW.md](../../WORKFLOW.md) for multi-phase, multi-provider work. The
+core idea: **feature flags decouple *merging* from *activating*, which makes every
+integration safe to land and trivial to roll back.**
+
+## Branching per phase
+| Phase | Milestone | Branches |
+|-------|-----------|----------|
+| 1 Provider abstraction + Google | `v1.x` | `epic/provider-platform` → `feature/email-provider-interface` · `feature/gmail-adapter` · `feature/oauth-login-users` |
+| 2 Microsoft | `v1.x` | `feature/outlook-provider` · `feature/microsoft-oauth` |
+| 3 Unified model + sync | `v1.x` | `feature/normalised-email-model` · `feature/sync-orchestrator` · `feature/data-migration` |
+| 4 Settings | `v1.x` | `feature/settings-store` · `feature/feature-flags-engine` |
+| 5 Admin | `v1.x` | `feature/admin-api` · `feature/admin-ui` · `feature/audit-log` |
+| 6 AI layer | `v1.x` | `feature/ai-abstraction-ext` · `feature/ai-analyzers` · `feature/ai-flag-gating` |
+- `feature/* → develop` (squash), `develop → main` (merge commit) — as established. Epics are
+  tracked by milestone/label; features integrate continuously (no long epic branch).
+
+## PR structure per provider integration
+Each provider is a self-contained PR set that lands **dark**:
+1. **Adapter PR** — `IEmailProvider` impl + normaliser + unit tests (mocked provider).
+2. **Auth PR** — OAuth login/link for that provider.
+3. **Enablement PR** — register in `ProviderFactory` + seed `provider.<x>` flag **OFF**.
+4. **Activation** — flip the flag on in staging → verify end-to-end → roll out in prod.
+- **PR checklist adds:** provider behind a flag (off by default) · normaliser tests · token
+  encryption verified · no Domain leakage · docs updated.
+
+## Feature flags prevent breaking changes
+- Merge = code present but **inert** until its flag is on. So half-finished providers/AI can
+  live on `main` safely; CI stays green; no long-lived divergence.
+- AI ships behind `ai.*`; providers behind `provider.*`; risky changes behind their own flag.
+
+## Rollback strategy (per provider / per feature)
+| Level | Action | Speed |
+|-------|--------|-------|
+| **Flag** (first resort) | Admin flips `provider.<x>` / `ai.<x>` **off** | **Instant, no deploy** — feature disappears, existing data untouched |
+| **Deploy** | Redeploy the previous **tag** (`vX.Y.Z-1`) | Minutes (pipeline) |
+| **Revert** | `git revert` the PR → PR → merge → deploy | Minutes–hours |
+| **Data** | Provider accounts are isolated; disabling a provider **pauses** its sync — no destructive change to migrate back | Safe by design |
+- Because providers are isolated and flag-gated, a bad integration **never blocks the others**
+  and never requires a risky data rollback.
+
+## Release milestones
+- Cut a tag when a phase reaches its exit criteria (`deploy-prod.yml` fires on `v*`).
+- Suggested: `v1.1` provider platform + Google · `v1.2` +Outlook · `v1.3` unified sync ·
+  `v1.4` settings+admin · `v1.5` AI layer — folded into the blueprint roadmap ([../09](../09-roadmap.md)).
+
+## Docs alongside code
+Every feature PR updates the relevant `multi-provider/*` doc + `CHANGELOG.md`; on approval the
+design docs graduate into living `docs/` references (provider system, settings, admin, security).

docs/discovery/multi-provider/11-risk-analysis.md

@@ -0,0 +1,31 @@
+# 11 — Risk Analysis
+
+Risks specific to the multi-provider + multi-user + admin evolution (the blueprint-wide risks
+are in [../11](../11-risks-and-future.md)). L/I = Likelihood/Impact (H/M/L).
+
+| # | Risk | L | I | Mitigation |
+|---|------|---|---|------------|
+| M1 | **Account-linking hijack** (attach someone's mailbox to your user) | L | H | Must authenticate as the target user; unique `(provider, sub)`; explicit "already owned" block; audited ([02](02-auth-and-signin.md)) |
+| M2 | **Cross-user data leakage** (multi-user IDOR) | M | H | EF global query filters on `user_id`/`account_id`; policy-based authz; no admin route reads user mail; tests for isolation |
+| M3 | **OAuth token theft / exposure** | L | H | Encrypted at rest (Data Protection); never logged or sent to browser; rotation; `bytea` ciphertext |
+| M4 | **Provider quirks break sync** (Graph delta resets, IMAP `UIDVALIDITY` change, Gmail history gaps) | M | M | Per-provider cursor handling with **full-resync fallback**; idempotent upserts; capability flags; account flips to needs-attention, never crashes |
+| M5 | **Migration corrupts existing Gmail data** | L | H | Additive-only schema; backfill is idempotent; maintenance window; tested on a staging copy; reversible ([12](12-migration-guide.md)) |
+| M6 | **RBAC bug grants Member admin powers** | L | H | Server-side policies (not UI); admin-only role changes audited + force session refresh; can't demote last Admin; authz tests |
+| M7 | **Feature-flag misconfiguration** (AI/provider on when not ready) | M | M | Flags default **off/fail-closed**; land dark; enable in staging first; audited toggles; rollout by role |
+| M8 | **AI gate bypass** (feature runs when disabled) | L | M | Single `IAiGate` chokepoint before any AI call; Null providers; capability checks; no direct provider refs in core |
+| M9 | **Session/CSRF weaknesses** across new surfaces | M | M | Server-side sessions, id rotation, SameSite + CSRF token, revoke-all, short idle expiry |
+| M10 | **Admin abuse / mistake** (mass suspend, wrong flag) | M | M | Step-up confirm + rate-limit + full audit trail + reversible flags |
+| M11 | **Scaling: millions of msgs × multiple accounts** | M | M | Denormalised `user_id`, keyset pagination, GIN/trgm/HNSW indexes, optional `account_id` partitioning, per-account sync throttling |
+| M12 | **Provider OAuth app setup burden** (separate Google + Microsoft app registrations, redirect URIs, verification) | M | L | Documented setup per provider; providers flag-gated so an unconfigured one is simply hidden |
+| M13 | **Scope/verification friction** (Google restricted scopes, MS admin consent) | M | M | Least-privilege scopes; document the consent/verification path; self-host uses the operator's own OAuth apps |
+
+## Top watch-items
+- **M2 (isolation)** and **M6 (RBAC)** — the two ways multi-user can go wrong; both mitigated
+  by server-side authz + query filters + isolation tests as a release gate.
+- **M4 (provider sync quirks)** — the most likely *operational* pain; the full-resync fallback
+  and per-account isolation contain it.
+
+## Overall
+The flag-gated, additive, isolated design makes this evolution **low-blast-radius**: each
+provider and the AI layer land dark and roll back by flag, migrations are additive, and the
+security model closes the new multi-user gaps. Proceed **phase by phase behind flags**.

docs/discovery/multi-provider/12-migration-guide.md

@@ -0,0 +1,52 @@
+# 12 — Migration Guide (Part 10)
+
+Moving the current **single-account Gmail** app to the **multi-provider, multi-user** model
+— **additive and reversible**, no destructive step. Executed as ordered EF Core migrations +
+idempotent backfills, behind a short maintenance window.
+
+## Principles
+- **Additive first:** create new tables/columns before moving data; keep old columns until parity is verified.
+- **Idempotent backfills:** safe to re-run; keyed on stable ids.
+- **Flag-gated cutover:** the new sign-in/model activates behind flags; the old path stays until removed.
+- **Reversible:** each step has a documented rollback; no data is deleted during migration.
+
+## Step-by-step
+1. **Schema (additive migration)**
+   - Create `users`, `accounts`, `provider_tokens`, `sessions`, `user_settings`,
+     `system_settings`, `feature_flags`, `audit_logs`.
+   - Add `account_id`, `user_id` (nullable) to the current email/thread tables; widen
+     `search_vector`; add nullable `embedding vector(768)` + `pgvector` extension.
+2. **Identity backfill**
+   - For the existing operator/user, create a `users` row; mark the **first user = Admin**.
+   - Create one **`Google` `account`** per existing identity (`is_login_identity=true`); move
+     current encrypted Gmail tokens → `provider_tokens`.
+3. **Email backfill**
+   - Set `account_id`/`user_id` on all existing `Email`/thread rows to the default Google account.
+   - Regenerate the widened `search_vector`; leave `embedding` null (backfilled later by the AI phase).
+   - Enforce the new unique keys `(account_id, provider_message_id)` / `(account_id, provider_thread_id)`.
+4. **Config seed**
+   - Seed `feature_flags`: `provider.google=on`, `provider.microsoft=off`, `provider.imap=off`,
+     `ai.enabled` = derived from the current `Ai:Mode` (Disabled→off), `ai.*`=off, `maintenance.*`=off.
+   - Create `system_settings` singleton; create `user_settings` from any existing per-user prefs (else defaults).
+5. **Cutover**
+   - Enable the new OAuth-as-login + unified sync behind their flags; verify on **staging** first
+     (the pipeline we built), then production via a tagged release.
+6. **Cleanup (later, separate migration)**
+   - Once parity is confirmed in production, drop obsolete columns/paths. Not part of the cutover.
+
+## Verification checklist
+- Existing user signs in via Google → lands on their mail unchanged.
+- Email counts match pre/post; search returns identical results for sample queries.
+- Tokens decrypt and refresh; sync resumes from the correct cursor.
+- No cross-user rows visible (isolation test).
+
+## Rollback
+- **Pre-cutover:** additive changes are inert → simply don't flip the flags; drop new tables if aborting.
+- **Post-cutover issue:** flip flags off / redeploy previous **tag**; old columns still present →
+  the legacy path still works. No data was deleted, so no data rollback is needed.
+
+## Provider-app prerequisites (operator setup)
+- **Google:** OAuth client (existing) + redirect `/(…)/signin/google`; scopes `gmail.readonly gmail.modify`.
+- **Microsoft:** register an Entra app; redirect `/(…)/signin/microsoft`; scopes `Mail.Read Mail.ReadWrite offline_access`; admin consent if required.
+- **IMAP (future):** per-account host/credentials; validated against the SSRF allowlist.
+- Because providers are flag-gated, an unconfigured provider is simply hidden — configure, then enable.

docs/discovery/multi-provider/README.md

@@ -0,0 +1,48 @@
+# Multi-Provider Email Platform + Admin/Settings — Design
+
+**Design + architecture phase. No implementation until approved.**
+
+Evolves InboxIntel from a single-account Gmail tool into a **multi-provider platform**
+(Gmail · Outlook/Graph · future IMAP) for a **small self-hosted team**, with settings,
+feature flags, and an admin panel. Extends — does not discard — the
+[discovery blueprint](../README.md).
+
+## Locked decisions (from interview)
+1. **Tenancy:** **small team, self-hosted, one org.** Roles = **Admin / Member**. Shared
+   system settings + feature flags; each member's mail is private to them. No multi-tenant
+   org table (one implicit org); the model stays extensible to multi-org later.
+2. **App identity = provider OAuth.** The first Google/Microsoft sign-in **creates/authenticates
+   the InboxIntel user**; additional mailboxes **link** to that same user. **No passwords stored.**
+
+## How this reshapes the blueprint (the "review" deltas)
+| Blueprint assumption | New reality |
+|----------------------|-------------|
+| Single-user, local-first | **Multi-user (small team)** with Admin/Member RBAC + admin panel |
+| Gmail-centric `Email`/`Sender` | **Account-scoped, provider-normalised** model (`IEmailProvider`) |
+| AI gated by `Ai:Mode` + user pref | **AI gated by system feature flag → user pref → capability** (flag wins) |
+| One implicit mailbox | **N provider accounts per user** (`accounts` table + per-account sync cursors) |
+| Sync = `GmailSyncWorker` | **Provider-agnostic sync orchestrator** dispatching to provider adapters |
+
+These deltas will be back-ported into main-blueprint docs [08](../08-technical-architecture.md)
+and [09](../09-roadmap.md) when this design is approved.
+
+## Documents
+| # | Doc | Covers (brief part) | Status |
+|---|-----|---------------------|--------|
+| 01 | [Provider Abstraction](01-provider-abstraction.md) | Part 1 | ✅ draft |
+| 02 | [Auth & Sign-in](02-auth-and-signin.md) | Part 2 | ✅ draft |
+| 03 | [Database Design](03-database-design.md) | Part 6 | ✅ draft |
+| 04 | [Settings & Feature Flags](04-settings-and-flags.md) | Part 3 | ✅ draft |
+| 05 | [Admin System](05-admin-system.md) | Part 4 | ✅ draft |
+| 06 | [Security Model](06-security-model.md) | Part 5 | ✅ draft |
+| 07 | [UX Flows](07-ux-flows.md) | Part 7 | ✅ draft |
+| 08 | [AI Feature-Flag Integration](08-ai-feature-flags.md) | Part 8 | ✅ draft |
+| 09 | [Implementation Plan](09-implementation-plan.md) | Part 9 | ✅ draft |
+| 10 | [Git Workflow](10-git-workflow.md) | Part 11 | ✅ draft |
+| 11 | [Risk Analysis](11-risk-analysis.md) | output | ✅ draft |
+| 12 | [Migration Guide](12-migration-guide.md) | Part 10 | ✅ draft |
+
+## Non-negotiables carried forward
+Provider logic **never leaks into Domain** · search works **across all a user's accounts**
+· emails stored in **one unified format** · **AI never required** for core function ·
+tokens **encrypted at rest** · admin actions **audit-logged**.

frontend/index.html

@@ -5,16 +5,8 @@
     <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
     <meta name="viewport" content="width=device-width, initial-scale=1.0" />
     <title>InboxIntel — Gmail analytics & cleanup</title>
-    <script>
+    <!-- Theme applied before first paint; external file so CSP can use script-src 'self'. -->
-      // Apply the saved theme before first paint to avoid a flash of the wrong mode.
+    <script src="/theme-init.js"></script>
-      (function () {
-        try {
-          var t = localStorage.getItem('ii:theme');
-          if (!t) t = matchMedia('(prefers-color-scheme: dark)').matches ? 'dark' : 'light';
-          if (t === 'dark') document.documentElement.classList.add('dark');
-        } catch (e) {}
-      })();
-    </script>
   </head>
   <body>
     <div id="root"></div>

frontend/nginx.conf

@@ -4,6 +4,19 @@ server {
     root /usr/share/nginx/html;
     index index.html;
 
+    # AUDIT M-2: security headers on the SPA. script-src 'self' works because the theme
+    # bootstrap lives in /theme-init.js (no inline scripts); style-src needs 'unsafe-inline'
+    # for React/Chart.js/grid-layout inline style attributes (low risk with script-src locked).
+    add_header Content-Security-Policy "default-src 'self'; script-src 'self'; style-src 'self' 'unsafe-inline'; img-src 'self' data:; font-src 'self'; connect-src 'self'; object-src 'none'; base-uri 'self'; form-action 'self'; frame-ancestors 'none'" always;
+    add_header X-Content-Type-Options "nosniff" always;
+    add_header X-Frame-Options "DENY" always;
+    add_header Referrer-Policy "no-referrer" always;
+
+    # Compress the SPA bundle (AUDIT perf note: ~680 KB JS).
+    gzip on;
+    gzip_types text/css application/javascript application/json image/svg+xml;
+    gzip_min_length 1024;
+
     # SPA fallback.
     location / {
         try_files $uri $uri/ /index.html;

frontend/package.json

@@ -9,6 +9,7 @@
     "preview": "vite preview --host"
   },
   "dependencies": {
+    "@fontsource-variable/inter": "^5.2.8",
     "@radix-ui/react-avatar": "^1.2.1",
     "@radix-ui/react-checkbox": "^1.3.6",
     "@radix-ui/react-dialog": "^1.1.18",
@@ -34,10 +35,10 @@
     "tailwind-merge": "^3.6.0"
   },
   "devDependencies": {
-    "@vitejs/plugin-react": "^4.3.1",
+    "@vitejs/plugin-react": "^6.0.3",
     "autoprefixer": "^10.5.2",
     "postcss": "^8.5.16",
     "tailwindcss": "^3.4.19",
-    "vite": "^5.3.1"
+    "vite": "^8.1.2"
   }
 }

frontend/public/theme-init.js

@@ -0,0 +1,9 @@
+// Applied before first paint to avoid a flash of the wrong theme. Lives in a file (not
+// inline) so the SPA can ship a CSP with script-src 'self' (AUDIT M-2).
+(function () {
+  try {
+    // Dark-first: default new users to dark unless they've chosen light.
+    var t = localStorage.getItem('ii:theme') || 'dark';
+    if (t === 'dark') document.documentElement.classList.add('dark');
+  } catch (e) {}
+})();

frontend/src/api/client.js

@@ -38,7 +38,6 @@ export const AnalyticsApi = {
   health: () => api.get('/analytics/health').then((r) => r.data),
   topSenders: (take = 20) => api.get(`/analytics/top-senders?take=${take}`).then((r) => r.data),
   volume: (days = 90) => api.get(`/analytics/volume?days=${days}`).then((r) => r.data),
-  heatmap: () => api.get('/analytics/heatmap').then((r) => r.data),
   categoryHeatmap: () => api.get('/analytics/category-heatmap').then((r) => r.data),
   attachments: () => api.get('/analytics/attachments').then((r) => r.data),
   sidebarCounts: () => api.get('/analytics/sidebar-counts').then((r) => r.data),
@@ -94,13 +93,22 @@ function folderToRequest(slug, page, pageSize) {
     case 'starred':   return { ...base, isStarred: true };
     case 'sent':      return { ...base, gmailLabel: 'SENT' };
     case 'drafts':    return { ...base, gmailLabel: 'DRAFT' };
-    case 'archive':   return { ...base, isInInbox: false, isTrashed: false };
+    // Archive = filed away: not in inbox, not trashed, and NOT Sent/Spam/Draft/Chat.
+    // Excluding those labels stops sent mail leaking into Archive.
+    case 'archive':   return { ...base, isInInbox: false, isTrashed: false, excludeGmailLabels: ['SENT', 'DRAFT', 'SPAM', 'TRASH', 'CHAT'] };
     case 'spam':      return { ...base, gmailLabel: 'SPAM' };
     case 'trash':     return { ...base, isTrashed: true };
+    // Pinned = Gmail's "Important" marker (a real per-message flag), not "everything".
+    case 'pinned':    return { ...base, isImportant: true };
+    // Read Later = only emails the user explicitly flagged (local marker).
+    case 'readlater': return { ...base, isReadLater: true };
+    // Unlabelled = emails not filed under any user-created label.
+    case 'unlabelled': return { ...base, hasUserLabels: false };
     // ── Special filters ──
     case 'large':     return { ...base, minSizeBytes: 5_000_000 };
+    // Old Mail = strictly older than 6 months.
     case 'old': {
-      const d = new Date(); d.setFullYear(d.getFullYear() - 1);
+      const d = new Date(); d.setMonth(d.getMonth() - 6);
       return { ...base, to: d.toISOString().slice(0, 10) };
     }
     // ── Smart folders ──

frontend/src/components/BulkToolbar.jsx

@@ -1,5 +1,5 @@
 import { useState } from 'react';
-import { MailOpen, Mail, Star, Archive, Trash2 } from 'lucide-react';
+import { MailOpen, Mail, Star, Archive, Trash2, X } from 'lucide-react';
 import { BulkApi } from '../api/client.js';
 import {
   Button, useToast,
@@ -49,25 +49,54 @@ export default function BulkToolbar({ selectedIds, onDone, onClear }) {
   };
 
   return (
-    <div className="flex flex-wrap items-center gap-2 rounded-lg border border-border bg-card px-3 py-2 shadow-sm">
+    <div
-      <span className="text-sm font-medium">{count} selected</span>
+      role="toolbar"
-      <div className="flex-1" />
+      aria-label={`${count} email${count === 1 ? '' : 's'} selected`}
-      <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.markRead, 'read', 'Marked read')}>
+      className="flex flex-wrap items-center gap-x-3 gap-y-2 rounded-lg border border-border bg-card px-3 py-2 shadow-sm"
-        <MailOpen /> Read
+    >
-      </Button>
+      {/* Selection count — primary emphasis so it reads first. */}
-      <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.markUnread, 'unread', 'Marked unread')}>
+      <div className="flex items-center gap-2">
-        <Mail /> Unread
+        <span
-      </Button>
+          aria-hidden="true"
-      <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.star, 'star', 'Starred')}>
+          className="inline-flex h-7 min-w-7 items-center justify-center rounded-full bg-primary px-2 text-sm font-semibold tabular-nums text-primary-foreground"
-        <Star /> Star
+        >
-      </Button>
+          {count}
-      <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.archive, 'archive', 'Archived')}>
+        </span>
-        <Archive /> Archive
+        <span className="text-sm font-medium text-foreground">
-      </Button>
+          selected
-      <Button variant="danger-outline" size="sm" disabled={busy} onClick={() => setConfirmTrash(true)}>
+        </span>
-        <Trash2 /> Trash
+        {/* Obvious clear-selection affordance, kept next to the count. */}
-      </Button>
+        <Button
-      <Button variant="ghost" size="sm" disabled={busy} onClick={onClear}>Cancel</Button>
+          variant="ghost"
+          size="icon-sm"
+          disabled={busy}
+          onClick={onClear}
+          aria-label="Clear selection"
+          title="Clear selection"
+        >
+          <X />
+        </Button>
+      </div>
+
+      <div className="flex-1 basis-full sm:basis-0" />
+
+      <div className="flex flex-wrap items-center gap-2">
+        <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.markRead, 'read', 'Marked read')}>
+          <MailOpen /> Read
+        </Button>
+        <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.markUnread, 'unread', 'Marked unread')}>
+          <Mail /> Unread
+        </Button>
+        <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.star, 'star', 'Starred')}>
+          <Star /> Star
+        </Button>
+        <Button variant="ghost" size="sm" disabled={busy} onClick={() => apply(BulkApi.archive, 'archive', 'Archived')}>
+          <Archive /> Archive
+        </Button>
+        <Button variant="danger-outline" size="sm" disabled={busy} onClick={() => setConfirmTrash(true)}>
+          <Trash2 /> Trash
+        </Button>
+      </div>
 
       <Dialog open={confirmTrash} onOpenChange={(o) => !busy && setConfirmTrash(o)}>
         <DialogContent>

frontend/src/components/EmailDetail.jsx

@@ -0,0 +1,115 @@
+import { useEffect, useState } from 'react';
+import { EmailApi } from '../api/client.js';
+
+// ── helpers ──────────────────────────────────────────────────────────────────
+
+const fmtSize = (b) => {
+  if (!b) return '';
+  if (b < 1024) return `${b} B`;
+  if (b < 1048576) return `${(b / 1024).toFixed(0)} KB`;
+  return `${(b / 1048576).toFixed(1)} MB`;
+};
+
+/**
+ * Shared reading-pane / full single-email view.
+ *
+ * Accepts either an `email` summary object (as emitted by the list rows) or a
+ * bare `emailId`. Fetches the full detail via EmailApi.get(id) and the AI
+ * summary lazily via EmailApi.summary(id). Renders subject, metadata, an AI
+ * summary button, the body, and per-email actions including "Open in Gmail".
+ *
+ * `onClose` collapses the pane.
+ */
+export default function EmailDetail({ email: summary, emailId, onClose }) {
+  const id = summary?.id ?? emailId;
+
+  const [detail, setDetail] = useState(null);
+  const [loading, setLoading] = useState(true);
+  const [aiSummary, setAiSummary] = useState(null);
+  const [aiLoading, setAiLoading] = useState(false);
+
+  useEffect(() => {
+    if (id == null) return;
+    let cancelled = false;
+    setDetail(null);
+    setLoading(true);
+    setAiSummary(null);
+    setAiLoading(false);
+    EmailApi.get(id)
+      .then((d) => { if (!cancelled) setDetail(d); })
+      .catch(() => { if (!cancelled) setDetail(null); })
+      .finally(() => { if (!cancelled) setLoading(false); });
+    return () => { cancelled = true; };
+  }, [id]);
+
+  const email = detail ?? summary ?? {};
+
+  const fetchAiSummary = () => {
+    if (id == null) return;
+    setAiLoading(true);
+    EmailApi.summary(id)
+      .then((r) => setAiSummary(r.summary))
+      .catch(() => setAiSummary(null))
+      .finally(() => setAiLoading(false));
+  };
+
+  const openInGmail = () => window.open(
+    `https://mail.google.com/mail/u/0/#all/${email.gmailMessageId}`,
+    '_blank', 'noopener,noreferrer'
+  );
+
+  return (
+    <div className="sv-detail">
+      <div className="sv-detail-topbar">
+        <button className="sv-close" onClick={onClose} aria-label="Close reading pane" title="Close">
+          ✕
+        </button>
+      </div>
+
+      <div className="sv-detail-header">
+        <div className="sv-detail-subject">{email.subject || '(no subject)'}</div>
+        <div className="sv-detail-meta">
+          <span>{email.senderDisplayName || email.senderAddress}</span>
+          {email.sentAtUtc && (
+            <>
+              <span className="sv-detail-sep">·</span>
+              <span>{new Date(email.sentAtUtc).toLocaleString()}</span>
+            </>
+          )}
+          {email.sizeEstimateBytes > 0 && (
+            <>
+              <span className="sv-detail-sep">·</span>
+              <span>{fmtSize(email.sizeEstimateBytes)}</span>
+            </>
+          )}
+        </div>
+      </div>
+
+      {!loading && (
+        <div className="sv-ai-summary">
+          {aiSummary != null ? (
+            <div className="sv-ai-summary-text">✨ {aiSummary}</div>
+          ) : (
+            <button className="btn-sm" onClick={fetchAiSummary} disabled={aiLoading}>
+              {aiLoading ? 'Summarising…' : '✨ AI summary'}
+            </button>
+          )}
+        </div>
+      )}
+
+      {loading && <div className="sv-body-loading muted">Loading message…</div>}
+
+      {!loading && detail?.bodyText && (
+        <div className="sv-body">{detail.bodyText}</div>
+      )}
+
+      {!loading && !detail?.bodyText && email.snippet && (
+        <div className="sv-snippet">{email.snippet}</div>
+      )}
+
+      <div className="sv-detail-actions">
+        <button className="btn-sm" onClick={openInGmail}>Open in Gmail ↗</button>
+      </div>
+    </div>
+  );
+}

frontend/src/components/EmailRow.jsx

@@ -1,5 +1,6 @@
 import { useState } from 'react';
 import { EmailApi } from '../api/client.js';
+import { Checkbox } from './ui/checkbox.jsx';
 
 const fmtDate = (iso) => {
   const d = new Date(iso);
@@ -16,7 +17,26 @@ const fmtSize = (b) => {
   return `${(b / 1048576).toFixed(1)} MB`;
 };
 
-export default function EmailRow({ email: initial, onRemove, selected, onToggleSelect, focused }) {
+// "Why this matched": the API wraps matched terms in U+E000/U+E001 sentinels (NOT HTML).
+// We tokenise and render the highlighted parts as <mark> React elements — React escapes
+// all text nodes, so untrusted email content can never inject markup (no dangerouslySetInnerHTML).
+const HL_START = String.fromCharCode(0xE000);
+const HL_STOP = String.fromCharCode(0xE001);
+const HL_RE = new RegExp(HL_START + '([\s\S]*?)' + HL_STOP, 'g');
+function renderHighlight(s) {
+  const out = [];
+  let last = 0, key = 0, m;
+  HL_RE.lastIndex = 0;
+  while ((m = HL_RE.exec(s)) !== null) {
+    if (m.index > last) out.push(s.slice(last, m.index));
+    out.push(<mark key={key++}>{m[1]}</mark>);
+    last = HL_RE.lastIndex;
+  }
+  if (last < s.length) out.push(s.slice(last));
+  return out;
+}
+
+export default function EmailRow({ email: initial, onRemove, selected, onToggleSelect, focused, onOpen }) {
   const [email, setEmail] = useState(initial);
   const [acting, setActing] = useState(false);
 
@@ -69,12 +89,16 @@ export default function EmailRow({ email: initial, onRemove, selected, onToggleS
   return (
     <tr
       className={`email-row${email.isUnread ? ' email-row--unread' : ''}${acting ? ' email-row--acting' : ''}${focused ? ' email-row--focused' : ''}`}
-      onClick={openInGmail}
+      onClick={() => onOpen ? onOpen(email) : openInGmail()}
-      title="Open in Gmail"
+      title={onOpen ? 'Open' : 'Open in Gmail'}
     >
       {onToggleSelect && (
         <td className="el-select" onClick={(e) => e.stopPropagation()}>
-          <input type="checkbox" checked={!!selected} onChange={() => onToggleSelect(email.id)} />
+          <Checkbox
+            checked={!!selected}
+            onCheckedChange={() => onToggleSelect(email.id)}
+            aria-label={selected ? 'Deselect email' : 'Select email'}
+          />
         </td>
       )}
       <td className="el-unread">{email.isUnread && <span className="unread-dot" />}</td>
@@ -83,7 +107,9 @@ export default function EmailRow({ email: initial, onRemove, selected, onToggleS
       </td>
       <td className="el-subject">
         <span className="el-subj-text">{email.subject || '(no subject)'}</span>
-        {email.snippet && <span className="el-snippet"> — {email.snippet}</span>}
+        {email.matchHighlight
+          ? <span className="el-snippet">{renderHighlight(email.matchHighlight)}</span>
+          : email.snippet && <span className="el-snippet">{email.snippet}</span>}
       </td>
       <td className="el-meta">
         {email.hasAttachments && <span className="el-attach" title="Has attachment">📎</span>}
@@ -113,6 +139,11 @@ export default function EmailRow({ email: initial, onRemove, selected, onToggleS
             disabled={email._unsubDone}
           >{email._unsubDone ? '✓' : '✉✕'}</button>
         )}
+        <button
+          className="action-btn"
+          title="Open in Gmail"
+          onClick={(e) => { e.stopPropagation(); openInGmail(); }}
+        >↗</button>
         <button
           className="action-btn action-btn--danger"
           title="Move to trash"

frontend/src/components/Layout.jsx

@@ -256,7 +256,7 @@ export default function Layout() {
           <Input
             ref={searchInputRef}
             type="search"
-            placeholder="Search…  from:  is:unread  has:attachment    ( / )"
+            placeholder="Search your inbox…"
             value={searchQuery}
             onChange={(e) => setSearchQuery(e.target.value)}
             aria-label="Search emails"

frontend/src/components/ui/checkbox.jsx

@@ -0,0 +1,43 @@
+import { forwardRef } from 'react';
+import * as CheckboxPrimitive from '@radix-ui/react-checkbox';
+import { Check } from 'lucide-react';
+import { cn } from '../../lib/utils.js';
+
+/**
+ * Accessible, design-token styled checkbox.
+ * Wraps Radix Checkbox so it is keyboard-accessible with a visible focus ring.
+ * Accepts `checked`, `onCheckedChange` (Radix) and, for convenience, `onChange`
+ * (called with a synthetic-ish `{ target: { checked } }`) so it can drop into
+ * places that previously used a bare <input type="checkbox">.
+ */
+const Checkbox = forwardRef(function Checkbox(
+  { className, onCheckedChange, onChange, ...props },
+  ref
+) {
+  const handleCheckedChange = (checked) => {
+    onCheckedChange?.(checked);
+    onChange?.({ target: { checked } });
+  };
+
+  return (
+    <CheckboxPrimitive.Root
+      ref={ref}
+      onCheckedChange={handleCheckedChange}
+      className={cn(
+        'peer inline-flex h-4 w-4 shrink-0 items-center justify-center rounded-[4px] border border-border bg-card transition-colors',
+        'focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-offset-1 focus-visible:ring-offset-background',
+        'hover:border-primary/60',
+        'disabled:cursor-not-allowed disabled:opacity-50',
+        'data-[state=checked]:border-primary data-[state=checked]:bg-primary data-[state=checked]:text-primary-foreground',
+        className
+      )}
+      {...props}
+    >
+      <CheckboxPrimitive.Indicator className="flex items-center justify-center text-current">
+        <Check className="h-3 w-3" strokeWidth={3} />
+      </CheckboxPrimitive.Indicator>
+    </CheckboxPrimitive.Root>
+  );
+});
+
+export { Checkbox };

frontend/src/components/ui/index.js

@@ -2,6 +2,7 @@
 export { Button, buttonVariants } from './button.jsx';
 export { Card, CardHeader, CardTitle, CardDescription, CardContent, CardFooter } from './card.jsx';
 export { Badge, badgeVariants } from './badge.jsx';
+export { Checkbox } from './checkbox.jsx';
 export { Input, Textarea } from './input.jsx';
 export { Switch } from './switch.jsx';
 export { Separator } from './separator.jsx';

frontend/src/components/widgets.jsx

@@ -1,13 +1,16 @@
 import { useEffect, useState } from 'react';
 import { useNavigate } from 'react-router-dom';
-import { Bar, Line, Doughnut } from 'react-chartjs-2';
+import { Line, Doughnut } from 'react-chartjs-2';
+import { Inbox } from 'lucide-react';
 import { AnalyticsApi } from '../api/client.js';
+import { Skeleton } from './ui/skeleton.jsx';
+import { EmptyState } from './ui/misc.jsx';
 import {
-  Chart as ChartJS, CategoryScale, LinearScale, BarElement, PointElement,
+  Chart as ChartJS, CategoryScale, LinearScale, PointElement,
   LineElement, ArcElement, Tooltip, Legend
 } from 'chart.js';
 
-ChartJS.register(CategoryScale, LinearScale, BarElement, PointElement, LineElement, ArcElement, Tooltip, Legend);
+ChartJS.register(CategoryScale, LinearScale, PointElement, LineElement, ArcElement, Tooltip, Legend);
 
 const fmtBytes = (b) => {
   if (!b) return '0 B';
@@ -77,32 +80,6 @@ export function VolumeWidget({ volume }) {
   );
 }
 
-export function HeatmapWidget({ heatmap }) {
-  if (!heatmap) return <Empty />;
-  const days = ['Sun', 'Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat'];
-  const max = Math.max(1, ...heatmap.map((c) => c.count));
-  const grid = {};
-  heatmap.forEach((c) => { grid[`${c.dayOfWeek}-${c.hour}`] = c.count; });
-  return (
-    <div className="widget">
-      <h3>Activity Heatmap</h3>
-      <div className="heatmap">
-        {days.map((d, dow) => (
-          <div className="hm-row" key={dow}>
-            <span className="hm-day">{d}</span>
-            {Array.from({ length: 24 }, (_, h) => {
-              const v = grid[`${dow}-${h}`] || 0;
-              return <span key={h} className="hm-cell" style={{ opacity: 0.1 + 0.9 * (v / max) }} title={`${d} ${h}:00 — ${v}`} />;
-            })}
-          </div>
-        ))}
-      </div>
-    </div>
-  );
-}
-
-const DOW = ['Sun', 'Mon', 'Tue', 'Wed', 'Thu', 'Fri', 'Sat'];
-
 // Maps EmailCategory enum name → folder slug or search query
 const CAT_SLUG = {
   Finance:          '/app/folder/finance',
@@ -135,41 +112,64 @@ const CAT_SLUG = {
   Unknown:          '/app/folder/allmail',
 };
 
-export function CategoryHeatmapWidget() {
+// "Emails by Category" — ranked horizontal bar list. Aggregates the
+// category×day-of-week cells into per-category totals; each bar links to
+// that category's folder via CAT_SLUG.
+export function CategoryBreakdownWidget() {
   const [cells, setCells] = useState(null);
   const navigate = useNavigate();
   useEffect(() => { AnalyticsApi.categoryHeatmap().then(setCells).catch(() => setCells([])); }, []);
 
-  if (!cells) return <div className="widget"><div className="muted">Loading…</div></div>;
+  if (!cells) {
-  if (cells.length === 0) return <div className="widget"><h3>Category Heatmap</h3><div className="muted">No data yet — run a sync.</div></div>;
+    return (
+      <div className="widget">
+        <h3>Emails by Category</h3>
+        <div className="cat-bars">
+          {Array.from({ length: 6 }, (_, i) => <Skeleton key={i} className="h-6 w-full" />)}
+        </div>
+      </div>
+    );
+  }
 
-  const categories = [...new Set(cells.map((c) => c.category))].sort();
+  const totals = {};
-  const max = Math.max(1, ...cells.map((c) => c.count));
+  cells.forEach((c) => { totals[c.category] = (totals[c.category] || 0) + c.count; });
-  const grid = {};
+  const ranked = Object.entries(totals)
-  cells.forEach((c) => { grid[`${c.category}-${c.dayOfWeek}`] = c.count; });
+    .map(([category, count]) => ({ category, count }))
+    .sort((a, b) => b.count - a.count)
+    .slice(0, 8);
+
+  if (ranked.length === 0) {
+    return (
+      <div className="widget">
+        <h3>Emails by Category</h3>
+        <EmptyState icon={Inbox} title="No data yet — run a sync" description="Category totals appear once your inbox has been synced." />
+      </div>
+    );
+  }
+
+  const max = Math.max(1, ...ranked.map((r) => r.count));
 
   return (
     <div className="widget">
-      <h3>Category Heatmap</h3>
+      <h3>Emails by Category</h3>
-      <div className="cat-heatmap">
+      <div className="cat-bars">
-        <div className="chm-row chm-head">
+        {ranked.map(({ category, count }) => {
-          <span className="chm-label" />
+          const dest = CAT_SLUG[category];
-          {DOW.map((d) => <span key={d} className="chm-col">{d}</span>)}
-        </div>
-        {categories.map((cat) => {
-          const dest = CAT_SLUG[cat];
           return (
-            <div className="chm-row" key={cat}>
+            <button
-              <span
+              type="button"
-                className={`chm-label${dest ? ' chm-label--link' : ''}`}
+              className={`cat-bar${dest ? ' cat-bar--link' : ''}`}
-                title={dest ? `View ${cat} emails` : cat}
+              key={category}
-                onClick={dest ? () => navigate(dest) : undefined}
+              disabled={!dest}
-              >{cat}</span>
+              title={dest ? `View ${category} emails` : category}
-              {DOW.map((_, dow) => {
+              onClick={dest ? () => navigate(dest) : undefined}
-                const v = grid[`${cat}-${dow}`] || 0;
+            >
-                return <span key={dow} className="chm-cell" style={{ opacity: 0.12 + 0.88 * (v / max) }} title={`${cat} · ${DOW[dow]} — ${v}`}>{v || ''}</span>;
+              <span className="cat-bar-label">{category}</span>
-              })}
+              <span className="cat-bar-track">
-            </div>
+                <span className="cat-bar-fill" style={{ width: `${(count / max) * 100}%` }} />
+              </span>
+              <span className="cat-bar-count">{count.toLocaleString()}</span>
+            </button>
           );
         })}
       </div>
@@ -196,4 +196,10 @@ export function StorageWidget({ bytes }) {
   return <StatCard label="Estimated Storage" value={fmtBytes(bytes)} />;
 }
 
-function Empty() { return <div className="widget"><div className="muted">No data yet — run a sync.</div></div>; }
+function Empty() {
+  return (
+    <div className="widget">
+      <EmptyState icon={Inbox} title="No data yet — run a sync" description="This widget populates after your inbox has been synced." />
+    </div>
+  );
+}

frontend/src/hooks/useListKeyboardNav.js

@@ -2,9 +2,15 @@ import { useEffect, useState } from 'react';
 import { BulkApi } from '../api/client.js';
 
 /// <summary>
-/// j/k move focus down/up the list, e archives the focused email, # (shift+3)
+/// Keyboard navigation for an email list. Shortcuts:
-/// trashes it. Ignored while an input/textarea/select has focus, or while
+///   j / ArrowDown  move focus down
-/// the "/" search shortcut is active, so typing is never hijacked.
+///   k / ArrowUp    move focus up
+///   e              archive the focused email
+///   u              mark the focused email unread
+///   #  (shift+3)   trash the focused email
+/// All shortcuts are ignored while an input/textarea/select (or any
+/// contenteditable) has focus, and modifier chords (Ctrl/Cmd/Alt) are left
+/// alone, so typing and browser/OS shortcuts are never hijacked.
 /// `onRemoved(id)` lets the caller drop the row from local state after a
 /// successful archive/trash.
 /// </summary>
@@ -12,31 +18,59 @@ export default function useListKeyboardNav(emails, onRemoved) {
   const [focusedId, setFocusedId] = useState(null);
 
   useEffect(() => {
+    const isEditable = (el) => {
+      if (!el) return false;
+      const tag = el.tagName;
+      return tag === 'INPUT' || tag === 'TEXTAREA' || tag === 'SELECT' || el.isContentEditable;
+    };
+
     const handler = async (e) => {
-      const tag = document.activeElement?.tagName;
+      // Never hijack typing or modifier chords (Ctrl+C, Cmd+K, Alt+…).
-      if (tag === 'INPUT' || tag === 'TEXTAREA' || tag === 'SELECT') return;
+      if (isEditable(document.activeElement)) return;
+      if (e.ctrlKey || e.metaKey || e.altKey) return;
       if (!emails.length) return;
 
       const idx = emails.findIndex((x) => x.id === focusedId);
 
-      if (e.key === 'j') {
+      switch (e.key) {
-        e.preventDefault();
+        case 'j':
-        const next = idx < 0 ? 0 : Math.min(idx + 1, emails.length - 1);
+        case 'ArrowDown': {
-        setFocusedId(emails[next].id);
+          e.preventDefault();
-      } else if (e.key === 'k') {
+          const next = idx < 0 ? 0 : Math.min(idx + 1, emails.length - 1);
-        e.preventDefault();
+          setFocusedId(emails[next].id);
-        const prev = idx < 0 ? 0 : Math.max(idx - 1, 0);
+          break;
-        setFocusedId(emails[prev].id);
+        }
-      } else if (e.key === 'e' && idx >= 0) {
+        case 'k':
-        e.preventDefault();
+        case 'ArrowUp': {
-        const id = emails[idx].id;
+          e.preventDefault();
-        await BulkApi.archive([id]);
+          const prev = idx < 0 ? 0 : Math.max(idx - 1, 0);
-        onRemoved(id);
+          setFocusedId(emails[prev].id);
-      } else if (e.key === '#' && idx >= 0) {
+          break;
-        e.preventDefault();
+        }
-        const id = emails[idx].id;
+        case 'e': {
-        await BulkApi.trash([id]);
+          if (idx < 0) break;
-        onRemoved(id);
+          e.preventDefault();
+          const id = emails[idx].id;
+          await BulkApi.archive([id]);
+          onRemoved(id);
+          break;
+        }
+        case 'u': {
+          if (idx < 0) break;
+          e.preventDefault();
+          await BulkApi.markUnread([emails[idx].id]);
+          break;
+        }
+        case '#': {
+          if (idx < 0) break;
+          e.preventDefault();
+          const id = emails[idx].id;
+          await BulkApi.trash([id]);
+          onRemoved(id);
+          break;
+        }
+        default:
+          break;
       }
     };
 

frontend/src/index.css

@@ -8,46 +8,55 @@
   two variables at runtime. See docs/specs/ui-overhaul.md.
 */
 @layer base {
+  /*
+    Design tokens — v2 (per docs/discovery/04). Dark-first, warm-leaning neutrals,
+    green brand accent derived from #3ba31f. Light is a genuine white (not grey).
+    Green is never the SOLE state signal (pair with icon/shape) — colour-blind-safe
+    by construction. Values are HSL triples (space-separated) for Tailwind's
+    `hsl(var(--x) / <alpha-value>)` mapping.
+  */
   :root {
-    /* Neutrals — warm-tinted slate (Notion-ish paper) */
+    /* Light — genuine white, warm off-white surfaces */
     --background: 0 0% 100%;
-    --foreground: 222 22% 12%;
+    --foreground: 30 10% 12%;
-    --card: 0 0% 100%;
+    --card: 40 33% 99%;
-    --muted: 220 16% 96%;
+    --muted: 40 24% 96%;
-    --muted-foreground: 220 9% 46%;
+    --muted-foreground: 35 8% 42%;
-    --border: 220 16% 90%;
+    --border: 38 18% 89%;
-    --input: 220 16% 90%;
+    --input: 38 18% 89%;
-    --ring: 245 75% 60%;
+    --ring: 110 62% 38%;
 
-    /* Brand accent — Indigo #5b5bf0 */
+    /* Brand accent — green (from #3ba31f), darkened for AA white-on-green */
-    --primary: 245 75% 59%;
+    --primary: 110 62% 33%;
     --primary-foreground: 0 0% 100%;
 
     /* Semantic */
-    --success: 152 56% 40%;
+    --success: 145 55% 38%;
-    --warning: 38 92% 50%;
+    --warning: 38 92% 45%;
-    --danger: 0 72% 51%;
+    --danger: 4 74% 50%;
     --danger-foreground: 0 0% 100%;
 
-    --radius: 0.625rem;
+    --radius: 0.5rem; /* lg 8px · md 6px · sm 4px */
   }
 
   .dark {
-    --background: 224 32% 9%;
+    /* Dark (default) — warm charcoal layers, NOT pure black */
-    --foreground: 220 18% 92%;
+    --background: 30 7% 10%;
-    --card: 224 28% 12%;
+    --foreground: 40 22% 93%;
-    --muted: 223 22% 17%;
+    --card: 30 7% 13%;
-    --muted-foreground: 220 12% 64%;
+    --muted: 30 6% 17%;
-    --border: 223 20% 20%;
+    --muted-foreground: 35 9% 64%;
-    --input: 223 20% 22%;
+    --border: 30 7% 22%;
-    --ring: 245 80% 66%;
+    --input: 30 7% 24%;
+    --ring: 110 50% 46%;
 
-    --primary: 245 80% 67%;
+    /* Brand accent — luminous green for dark surfaces */
-    --primary-foreground: 224 32% 9%;
+    --primary: 110 52% 42%;
+    --primary-foreground: 0 0% 100%;
 
-    --success: 152 50% 50%;
+    --success: 145 50% 48%;
-    --warning: 38 92% 58%;
+    --warning: 38 90% 56%;
-    --danger: 0 70% 60%;
+    --danger: 4 72% 58%;
     --danger-foreground: 0 0% 100%;
   }
 
@@ -55,10 +64,19 @@
     border-color: hsl(var(--border));
   }
 
+  /* Search "why this matched" highlight — subtle accent tint, not the default yellow. */
+  mark {
+    background: hsl(var(--primary) / 0.22);
+    color: inherit;
+    border-radius: 3px;
+    padding: 0 1px;
+  }
+
   body {
     @apply bg-background text-foreground antialiased;
-    font-family: ui-sans-serif, system-ui, -apple-system, 'Segoe UI', Roboto,
+    /* Inter (variable, self-hosted via @fontsource-variable/inter); system fallback. */
-      Helvetica, Arial, sans-serif;
+    font-family: 'Inter Variable', 'Inter', ui-sans-serif, system-ui, -apple-system,
+      'Segoe UI', Roboto, Helvetica, Arial, sans-serif;
   }
 
   /* Subtle, modern scrollbars that respect the theme. */
@@ -82,6 +100,29 @@
   }
 }
 
+@layer base {
+  /* Consistent, visible focus ring using the accent token (keyboard users). */
+  :focus-visible {
+    outline: 2px solid hsl(var(--ring));
+    outline-offset: 2px;
+  }
+
+  /*
+    Respect the OS "reduce motion" setting — near-instant, no large transitions.
+    Baseline accessibility (the deeper a11y pass is deferred; see the design brief).
+  */
+  @media (prefers-reduced-motion: reduce) {
+    *,
+    *::before,
+    *::after {
+      animation-duration: 0.01ms !important;
+      animation-iteration-count: 1 !important;
+      transition-duration: 0.01ms !important;
+      scroll-behavior: auto !important;
+    }
+  }
+}
+
 @layer utilities {
   .sr-only {
     position: absolute;

frontend/src/main.jsx

@@ -1,39 +1,54 @@
-import React from 'react';
+import React, { Suspense, lazy } from 'react';
 import ReactDOM from 'react-dom/client';
 import { BrowserRouter, Routes, Route, Navigate } from 'react-router-dom';
 import Landing from './pages/Landing.jsx';
-import Dashboard from './pages/Dashboard.jsx';
-import Senders from './pages/Senders.jsx';
-import Cleanup from './pages/Cleanup.jsx';
-import Unsubscribe from './pages/Unsubscribe.jsx';
-import FolderView from './pages/FolderView.jsx';
-import SearchResults from './pages/SearchResults.jsx';
 import Layout from './components/Layout.jsx';
 import { ToastProvider, TooltipProvider } from './components/ui';
+import '@fontsource-variable/inter';
 import './index.css';
 import './styles.css';
+import './split.css';
+
+// Route-level code splitting: each page loads its own chunk on first visit, so the
+// initial bundle no longer carries Chart.js / grid-layout / every page at once.
+// Landing + Layout stay eager (they're the first paint).
+const Dashboard = lazy(() => import('./pages/Dashboard.jsx'));
+const Senders = lazy(() => import('./pages/Senders.jsx'));
+const Cleanup = lazy(() => import('./pages/Cleanup.jsx'));
+const Unsubscribe = lazy(() => import('./pages/Unsubscribe.jsx'));
+const FolderView = lazy(() => import('./pages/FolderView.jsx'));
+const SearchResults = lazy(() => import('./pages/SearchResults.jsx'));
+const DesignSystem = lazy(() => import('./pages/DesignSystem.jsx'));
+
+// Minimal, theme-correct route fallback (skeleton-style, per the design system).
+const RouteFallback = () => (
+  <div className="p-6 text-sm text-muted-foreground" aria-busy="true">Loading…</div>
+);
 
 ReactDOM.createRoot(document.getElementById('root')).render(
   <React.StrictMode>
     <ToastProvider>
       <TooltipProvider delayDuration={200}>
         <BrowserRouter>
-          <Routes>
+          <Suspense fallback={<RouteFallback />}>
-        {/* Public landing page */}
+            <Routes>
-        <Route path="/" element={<Landing />} />
+              {/* Public landing page */}
+              <Route path="/" element={<Landing />} />
 
-        {/* Authenticated app */}
+              {/* Authenticated app */}
-        <Route path="/app" element={<Layout />}>
+              <Route path="/app" element={<Layout />}>
-          <Route index element={<Dashboard />} />
+                <Route index element={<Dashboard />} />
-          <Route path="senders" element={<Senders />} />
+                <Route path="senders" element={<Senders />} />
-          <Route path="cleanup" element={<Cleanup />} />
+                <Route path="cleanup" element={<Cleanup />} />
-          <Route path="unsubscribe" element={<Unsubscribe />} />
+                <Route path="unsubscribe" element={<Unsubscribe />} />
-          <Route path="folder/:slug" element={<FolderView />} />
+                <Route path="folder/:slug" element={<FolderView />} />
-          <Route path="search" element={<SearchResults />} />
+                <Route path="search" element={<SearchResults />} />
-        </Route>
+                <Route path="design" element={<DesignSystem />} />
+              </Route>
 
-            <Route path="*" element={<Navigate to="/" replace />} />
+              <Route path="*" element={<Navigate to="/" replace />} />
-          </Routes>
+            </Routes>
+          </Suspense>
         </BrowserRouter>
       </TooltipProvider>
     </ToastProvider>

frontend/src/pages/Dashboard.jsx

@@ -6,8 +6,9 @@ import { AnalyticsApi, LayoutApi, ExportApi, SyncApi } from '../api/client.js';
 import SyncSplash from '../components/SyncSplash.jsx';
 import {
   HealthWidget, StatCard, TopSendersWidget, VolumeWidget,
-  CategoryHeatmapWidget, AttachmentsWidget, StorageWidget
+  CategoryBreakdownWidget, AttachmentsWidget, StorageWidget
 } from '../components/widgets.jsx';
+import { Skeleton } from '../components/ui/skeleton.jsx';
 
 // Default grid geometry; overridden by the user's saved layout.
 const DEFAULT_LAYOUT = [
@@ -17,12 +18,23 @@ const DEFAULT_LAYOUT = [
   { i: 'storage', x: 7, y: 0, w: 2, h: 2 },
   { i: 'top-senders', x: 3, y: 2, w: 3, h: 5 },
   { i: 'volume', x: 6, y: 2, w: 6, h: 4 },
-  { i: 'category-heatmap', x: 0, y: 5, w: 6, h: 5 },
+  { i: 'category-breakdown', x: 0, y: 5, w: 6, h: 5 },
   { i: 'attachments', x: 6, y: 6, w: 4, h: 4 },
 ];
 
 const ALL_WIDGETS = DEFAULT_LAYOUT.map((l) => l.i);
 
+function WidgetSkeleton() {
+  return (
+    <div className="widget">
+      <Skeleton className="mb-3 h-4 w-1/3" />
+      <Skeleton className="mb-2 h-3 w-full" />
+      <Skeleton className="mb-2 h-3 w-5/6" />
+      <Skeleton className="h-3 w-2/3" />
+    </div>
+  );
+}
+
 export default function Dashboard() {
   const [data, setData] = useState(null);
   const [layout, setLayout] = useState(DEFAULT_LAYOUT);
@@ -87,15 +99,18 @@ export default function Dashboard() {
   const visibleLayout = useMemo(() => layout.filter((l) => !hidden.includes(l.i)), [layout, hidden]);
 
   const render = (key) => {
+    // category-breakdown self-fetches, so it renders regardless of dashboard load state.
+    if (key === 'category-breakdown') return <CategoryBreakdownWidget />;
+    // While the dashboard payload loads, show a skeleton placeholder per widget.
+    if (!data) return <WidgetSkeleton />;
     switch (key) {
-      case 'inbox-health': return <HealthWidget health={data?.health} />;
+      case 'inbox-health': return <HealthWidget health={data.health} />;
-      case 'total-emails': return <StatCard label="Total Emails" value={(data?.totalEmails ?? 0).toLocaleString()} to="/app/folder/allmail" />;
+      case 'total-emails': return <StatCard label="Total Emails" value={(data.totalEmails ?? 0).toLocaleString()} to="/app/folder/allmail" />;
-      case 'unread-emails': return <StatCard label="Unread" value={(data?.unreadEmails ?? 0).toLocaleString()} to="/app/folder/unread" />;
+      case 'unread-emails': return <StatCard label="Unread" value={(data.unreadEmails ?? 0).toLocaleString()} to="/app/folder/unread" />;
-      case 'storage': return <StorageWidget bytes={data?.storageEstimateBytes} />;
+      case 'storage': return <StorageWidget bytes={data.storageEstimateBytes} />;
-      case 'top-senders': return <TopSendersWidget senders={data?.topSenders} />;
+      case 'top-senders': return <TopSendersWidget senders={data.topSenders} />;
-      case 'volume': return <VolumeWidget volume={data?.volumeOverTime} />;
+      case 'volume': return <VolumeWidget volume={data.volumeOverTime} />;
-      case 'category-heatmap': return <CategoryHeatmapWidget />;
+      case 'attachments': return <AttachmentsWidget attachments={data.attachmentBreakdown} />;
-      case 'attachments': return <AttachmentsWidget attachments={data?.attachmentBreakdown} />;
       default: return null;
     }
   };

frontend/src/pages/DesignSystem.jsx

@@ -0,0 +1,177 @@
+import React from 'react';
+import {
+  Button, Badge, Input, Textarea, Switch, Skeleton, Separator,
+  Card, CardHeader, CardTitle, CardDescription, CardContent,
+  Tabs, TabsList, TabsTrigger, TabsContent, EmptyState, ThemeToggle,
+} from '../components/ui';
+
+/*
+  Design-system preview (v2). A living reference for the tokens + primitives so the
+  redesign can be visually verified in both themes. Route: /app/design.
+  Not linked in primary nav — it's a developer/design surface.
+*/
+
+const TOKENS = [
+  ['background', 'App background'],
+  ['card', 'Card / surface'],
+  ['muted', 'Muted surface'],
+  ['border', 'Border'],
+  ['primary', 'Brand (green)'],
+  ['success', 'Success'],
+  ['warning', 'Warning'],
+  ['danger', 'Danger'],
+];
+// Full literal class names so Tailwind's JIT includes them.
+const GREEN = [
+  'bg-green-50', 'bg-green-100', 'bg-green-200', 'bg-green-300', 'bg-green-400',
+  'bg-green-500', 'bg-green-600', 'bg-green-700', 'bg-green-800', 'bg-green-900',
+];
+const TYPE = [
+  ['text-3xl', '30px — Display'],
+  ['text-2xl', '24px — Heading'],
+  ['text-xl', '20px — Subheading'],
+  ['text-lg', '18px — Large'],
+  ['text-base', '16px — Body'],
+  ['text-sm', '14px — UI base'],
+  ['text-xs', '12px — Caption'],
+];
+
+function Swatch({ token, label }) {
+  return (
+    <div className="flex flex-col gap-1">
+      <div
+        className="h-14 w-full rounded-lg border"
+        style={{ background: `hsl(var(--${token}))` }}
+      />
+      <div className="text-xs font-medium">{label}</div>
+      <div className="text-xs text-muted-foreground">--{token}</div>
+    </div>
+  );
+}
+
+function Section({ title, children }) {
+  return (
+    <section className="mb-10">
+      <h2 className="mb-3 text-lg font-semibold">{title}</h2>
+      {children}
+    </section>
+  );
+}
+
+export default function DesignSystem() {
+  return (
+    <div className="mx-auto max-w-4xl p-2">
+      <div className="mb-8 flex items-center justify-between">
+        <div>
+          <h1 className="text-2xl font-semibold">Design system</h1>
+          <p className="text-sm text-muted-foreground">
+            v2 tokens &amp; primitives — green accent, warm neutrals, dark-first. Toggle the
+            theme to verify both.
+          </p>
+        </div>
+        <ThemeToggle />
+      </div>
+
+      <Section title="Semantic tokens">
+        <div className="grid grid-cols-2 gap-4 sm:grid-cols-4">
+          {TOKENS.map(([t, l]) => <Swatch key={t} token={t} label={l} />)}
+        </div>
+      </Section>
+
+      <Section title="Brand ramp (green, from #3ba31f)">
+        <div className="flex overflow-hidden rounded-lg border">
+          {GREEN.map((cls) => (
+            <div key={cls} className={`h-12 flex-1 ${cls}`} title={cls} />
+          ))}
+        </div>
+        <div className="mt-1 flex justify-between text-xs text-muted-foreground">
+          <span>50</span><span>500 (brand)</span><span>900</span>
+        </div>
+      </Section>
+
+      <Section title="Typography (Inter)">
+        <div className="space-y-2">
+          {TYPE.map(([cls, label]) => (
+            <div key={cls} className={`${cls} font-medium`}>
+              {label}
+            </div>
+          ))}
+        </div>
+      </Section>
+
+      <Section title="Buttons">
+        <div className="flex flex-wrap items-center gap-3">
+          <Button>Primary</Button>
+          <Button variant="secondary">Secondary</Button>
+          <Button variant="outline">Outline</Button>
+          <Button variant="ghost">Ghost</Button>
+          <Button variant="destructive">Danger</Button>
+          <Button disabled>Disabled</Button>
+        </div>
+      </Section>
+
+      <Section title="Badges">
+        <div className="flex flex-wrap items-center gap-3">
+          <Badge>Default</Badge>
+          <Badge variant="secondary">Secondary</Badge>
+          <Badge variant="outline">Outline</Badge>
+          <Badge variant="destructive">Danger</Badge>
+        </div>
+      </Section>
+
+      <Section title="Inputs">
+        <div className="grid max-w-md gap-3">
+          <Input placeholder="Search your inbox…" />
+          <Textarea placeholder="Write a reply…" rows={3} />
+          <label className="flex items-center gap-2 text-sm">
+            <Switch defaultChecked /> Use AI features
+          </label>
+        </div>
+      </Section>
+
+      <Section title="Card & tabs">
+        <Card className="max-w-md">
+          <CardHeader>
+            <CardTitle>Thread summary</CardTitle>
+            <CardDescription>AI-generated · local</CardDescription>
+          </CardHeader>
+          <CardContent>
+            <Tabs defaultValue="summary">
+              <TabsList>
+                <TabsTrigger value="summary">Summary</TabsTrigger>
+                <TabsTrigger value="actions">Actions</TabsTrigger>
+              </TabsList>
+              <TabsContent value="summary" className="pt-3 text-sm text-muted-foreground">
+                A concise overview of the conversation would appear here.
+              </TabsContent>
+              <TabsContent value="actions" className="pt-3 text-sm text-muted-foreground">
+                • Reply to the client · • Follow up in 3 days
+              </TabsContent>
+            </Tabs>
+          </CardContent>
+        </Card>
+      </Section>
+
+      <Section title="Loading & empty states">
+        <div className="grid gap-6 sm:grid-cols-2">
+          <div className="space-y-2">
+            <Skeleton className="h-4 w-3/4" />
+            <Skeleton className="h-4 w-1/2" />
+            <Skeleton className="h-4 w-2/3" />
+          </div>
+          <div className="rounded-lg border p-4">
+            <EmptyState
+              title="No results"
+              description="Try broader terms or clear a filter."
+            />
+          </div>
+        </div>
+      </Section>
+
+      <Separator className="my-8" />
+      <p className="text-xs text-muted-foreground">
+        InboxIntel design system · see docs/discovery/04-ux-redesign-and-design-system.md
+      </p>
+    </div>
+  );
+}

frontend/src/pages/FolderView.jsx

@@ -2,7 +2,9 @@ import { useEffect, useState, useCallback, useRef } from 'react';
 import { useParams } from 'react-router-dom';
 import { SearchApi } from '../api/client.js';
 import EmailRow from '../components/EmailRow.jsx';
+import EmailDetail from '../components/EmailDetail.jsx';
 import BulkToolbar from '../components/BulkToolbar.jsx';
+import { Skeleton, EmptyState } from '../components/ui';
 import useSelection from '../hooks/useSelection.js';
 import useListKeyboardNav from '../hooks/useListKeyboardNav.js';
 
@@ -35,6 +37,20 @@ const FOLDER_META = {
 
 const PAGE_SIZE = 50;
 
+function ListSkeleton({ rows = 6 }) {
+  return (
+    <div className="sv-skeleton-list" aria-hidden="true">
+      {Array.from({ length: rows }).map((_, i) => (
+        <div className="sv-skeleton-row" key={i}>
+          <Skeleton className="h-3 w-3 rounded-full" />
+          <Skeleton className="h-3 flex-1" />
+          <Skeleton className="h-3 w-16" />
+        </div>
+      ))}
+    </div>
+  );
+}
+
 
 export default function FolderView() {
   const { slug } = useParams();
@@ -46,6 +62,7 @@ export default function FolderView() {
   const [hasMore, setHasMore] = useState(true);
   const [loading, setLoading] = useState(false);
   const [error, setError] = useState(null);
+  const [selectedEmail, setSelectedEmail] = useState(null);
   const { selected, toggle, clear, removeIds, selectedIds } = useSelection();
   const focusedId = useListKeyboardNav(emails, (id) => setEmails((prev) => prev.filter((x) => x.id !== id)));
 
@@ -56,6 +73,7 @@ export default function FolderView() {
     setPage(1);
     setHasMore(true);
     setError(null);
+    setSelectedEmail(null);
     clear();
   }, [slug, clear]);
 
@@ -115,34 +133,57 @@ export default function FolderView() {
         }}
       />
 
-      {!loading && !error && emails.length === 0 && (
+      <div className={`sv-split${selectedEmail ? ' sv-split--open' : ''}`}>
-        <div className="fv-empty">No emails in this folder.</div>
+        <div className="sv-list-pane">
-      )}
+          {loading && emails.length === 0 && !error && <ListSkeleton />}
 
-      {emails.length > 0 && (
+          {!loading && !error && emails.length === 0 && (
-        <table className="email-list">
+            <EmptyState
-          <tbody>
+              title="No emails in this folder"
-            {emails.map((e) => (
+              description="Nothing here yet — try another folder or run a sync."
-              <EmailRow
+            />
-                key={e.id}
+          )}
-                email={e}
-                selected={selected.has(e.id)}
-                onToggleSelect={toggle}
-                focused={focusedId === e.id}
-                onRemove={(id) => setEmails((prev) => prev.filter((x) => x.id !== id))}
-              />
-            ))}
-          </tbody>
-        </table>
-      )}
 
-      {/* Sentinel — triggers next page load when scrolled into view */}
+          {emails.length > 0 && (
-      <div ref={sentinelRef} className="fv-sentinel" />
+            <table className="email-list">
+              <tbody>
+                {emails.map((e) => (
+                  <EmailRow
+                    key={e.id}
+                    email={e}
+                    selected={selected.has(e.id)}
+                    onToggleSelect={toggle}
+                    focused={focusedId === e.id}
+                    onOpen={(email) => setSelectedEmail(email)}
+                    onRemove={(id) => {
+                      setEmails((prev) => prev.filter((x) => x.id !== id));
+                      setSelectedEmail((cur) => (cur?.id === id ? null : cur));
+                    }}
+                  />
+                ))}
+              </tbody>
+            </table>
+          )}
 
-      {loading && <div className="fv-loading-more">Loading…</div>}
+          {/* Sentinel — triggers next page load when scrolled into view */}
-      {!hasMore && emails.length > 0 && (
+          <div ref={sentinelRef} className="fv-sentinel" />
-        <div className="fv-end">— {emails.length.toLocaleString()} emails —</div>
+
-      )}
+          {loading && emails.length > 0 && <div className="fv-loading-more">Loading…</div>}
+          {!hasMore && emails.length > 0 && (
+            <div className="fv-end">— {emails.length.toLocaleString()} emails —</div>
+          )}
+        </div>
+
+        {selectedEmail && (
+          <aside className="sv-reading-pane">
+            <EmailDetail
+              key={selectedEmail.id}
+              email={selectedEmail}
+              onClose={() => setSelectedEmail(null)}
+            />
+          </aside>
+        )}
+      </div>
     </div>
   );
 }

frontend/src/pages/SearchResults.jsx

@@ -2,13 +2,29 @@ import { useEffect, useState, useCallback, useRef } from 'react';
 import { useSearchParams } from 'react-router-dom';
 import { SearchApi } from '../api/client.js';
 import EmailRow from '../components/EmailRow.jsx';
+import EmailDetail from '../components/EmailDetail.jsx';
 import BulkToolbar from '../components/BulkToolbar.jsx';
+import { Skeleton, EmptyState } from '../components/ui';
 import useSelection from '../hooks/useSelection.js';
 import useListKeyboardNav from '../hooks/useListKeyboardNav.js';
 import useSavedSearches from '../hooks/useSavedSearches.js';
 
 const PAGE_SIZE = 50;
 
+function ListSkeleton({ rows = 6 }) {
+  return (
+    <div className="sv-skeleton-list" aria-hidden="true">
+      {Array.from({ length: rows }).map((_, i) => (
+        <div className="sv-skeleton-row" key={i}>
+          <Skeleton className="h-3 w-3 rounded-full" />
+          <Skeleton className="h-3 flex-1" />
+          <Skeleton className="h-3 w-16" />
+        </div>
+      ))}
+    </div>
+  );
+}
+
 
 export default function SearchResults() {
   const [searchParams] = useSearchParams();
@@ -20,6 +36,7 @@ export default function SearchResults() {
   const [hasMore, setHasMore] = useState(true);
   const [loading, setLoading] = useState(false);
   const [error, setError] = useState(null);
+  const [selectedEmail, setSelectedEmail] = useState(null);
   const { selected, toggle, clear, removeIds, selectedIds } = useSelection();
   const focusedId = useListKeyboardNav(emails, (id) => setEmails((prev) => prev.filter((x) => x.id !== id)));
   const { searches: savedSearches, add: addSavedSearch } = useSavedSearches();
@@ -36,6 +53,7 @@ export default function SearchResults() {
     setPage(1);
     setHasMore(true);
     setError(null);
+    setSelectedEmail(null);
     clear();
   }, [q, clear]);
 
@@ -83,11 +101,13 @@ export default function SearchResults() {
         )}
       </div>
 
-      {!q.trim() && <div className="fv-empty">Enter a search query above.</div>}
+      {!q.trim() && (
-      {error && <div className="fv-error">{error}</div>}
+        <EmptyState
-      {!loading && !error && q.trim() && emails.length === 0 && !hasMore && (
+          title="Search your mail"
-        <div className="fv-empty">No results for "{q}".</div>
+          description="Enter a search query above to find emails."
+        />
       )}
+      {error && <div className="fv-error">{error}</div>}
 
       <BulkToolbar
         selectedIds={selectedIds}
@@ -100,27 +120,56 @@ export default function SearchResults() {
         }}
       />
 
-      {emails.length > 0 && (
+      {q.trim() && (
-        <table className="email-list">
+        <div className={`sv-split${selectedEmail ? ' sv-split--open' : ''}`}>
-          <tbody>
+          <div className="sv-list-pane">
-            {emails.map((e) => (
+            {loading && emails.length === 0 && !error && <ListSkeleton />}
-              <EmailRow
-                key={e.id}
-                email={e}
-                selected={selected.has(e.id)}
-                onToggleSelect={toggle}
-                focused={focusedId === e.id}
-                onRemove={(id) => setEmails((prev) => prev.filter((x) => x.id !== id))}
-              />
-            ))}
-          </tbody>
-        </table>
-      )}
 
-      <div ref={sentinelRef} className="fv-sentinel" />
+            {!loading && !error && emails.length === 0 && !hasMore && (
-      {loading && <div className="fv-loading-more">Loading…</div>}
+              <EmptyState
-      {!hasMore && emails.length > 0 && (
+                title={`No results for "${q}"`}
-        <div className="fv-end">— {emails.length.toLocaleString()} results —</div>
+                description="Try a different search term or filter."
+              />
+            )}
+
+            {emails.length > 0 && (
+              <table className="email-list">
+                <tbody>
+                  {emails.map((e) => (
+                    <EmailRow
+                      key={e.id}
+                      email={e}
+                      selected={selected.has(e.id)}
+                      onToggleSelect={toggle}
+                      focused={focusedId === e.id}
+                      onOpen={(email) => setSelectedEmail(email)}
+                      onRemove={(id) => {
+                        setEmails((prev) => prev.filter((x) => x.id !== id));
+                        setSelectedEmail((cur) => (cur?.id === id ? null : cur));
+                      }}
+                    />
+                  ))}
+                </tbody>
+              </table>
+            )}
+
+            <div ref={sentinelRef} className="fv-sentinel" />
+            {loading && emails.length > 0 && <div className="fv-loading-more">Loading…</div>}
+            {!hasMore && emails.length > 0 && (
+              <div className="fv-end">— {emails.length.toLocaleString()} results —</div>
+            )}
+          </div>
+
+          {selectedEmail && (
+            <aside className="sv-reading-pane">
+              <EmailDetail
+                key={selectedEmail.id}
+                email={selectedEmail}
+                onClose={() => setSelectedEmail(null)}
+              />
+            </aside>
+          )}
+        </div>
       )}
     </div>
   );

frontend/src/split.css

@@ -0,0 +1,148 @@
+/* ── Split-view reading pane ────────────────────────────────────────────────
+ * Owned by Unit 1. Gmail/Outlook-style master-detail: the email list stays on
+ * the left, a collapsible + horizontally resizable reading pane sits on the
+ * right. Below 768px the pane overlays the list (mobile stack) instead of
+ * squishing the columns side-by-side.
+ *
+ * These pages (.folder-view) are styled from styles.css, so we reference its
+ * legacy CSS variables (--panel, --panel-2, --text, --muted, --accent) with
+ * hardcoded hex fallbacks. The modern index.css tokens are stored as raw HSL
+ * channel triplets and only work via hsl(var(--x)), so they are NOT used bare
+ * here.
+ * ------------------------------------------------------------------------- */
+
+.sv-split {
+  display: flex;
+  align-items: stretch;
+  gap: 0;
+  width: 100%;
+}
+
+/* Left column — the list. Flexes to fill remaining space and scrolls itself. */
+.sv-list-pane {
+  flex: 1 1 auto;
+  min-width: 0;
+  overflow-x: auto;
+}
+
+/* Right column — the reading pane. Resizable via the native handle; the width
+ * gives it a sensible default, and the user can drag the bottom-right corner.
+ * `resize: horizontal` needs `overflow` other than visible. */
+.sv-reading-pane {
+  flex: 0 0 auto;
+  width: clamp(320px, 42%, 640px);
+  min-width: 300px;
+  max-width: 80vw;
+  border-left: 1px solid var(--panel-2, #222a3d);
+  overflow: auto;
+  resize: horizontal;
+  background: var(--panel, #1a2030);
+  align-self: stretch;
+}
+
+/* ── Reading pane inner content ─────────────────────────────────────────── */
+.sv-detail { padding: 18px 20px; }
+
+.sv-detail-topbar {
+  display: flex;
+  justify-content: flex-end;
+  margin-bottom: 6px;
+}
+.sv-close {
+  background: transparent;
+  border: 1px solid var(--panel-2, #222a3d);
+  color: var(--muted, #8b93a7);
+  border-radius: 6px;
+  width: 28px;
+  height: 28px;
+  font-size: 13px;
+  line-height: 1;
+  cursor: pointer;
+}
+.sv-close:hover {
+  background: var(--panel-2, #222a3d);
+  color: var(--text, #e6e9f0);
+}
+
+.sv-detail-header { margin-bottom: 14px; }
+.sv-detail-subject {
+  font-size: 18px;
+  font-weight: 700;
+  margin-bottom: 6px;
+  color: var(--text, #e6e9f0);
+  word-break: break-word;
+}
+.sv-detail-meta {
+  font-size: 12px;
+  color: var(--muted, #8b93a7);
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+  align-items: center;
+}
+.sv-detail-sep { opacity: 0.4; }
+
+.sv-ai-summary { margin-bottom: 14px; }
+.sv-ai-summary-text {
+  background: var(--panel-2, #222a3d);
+  border: 1px solid var(--panel-2, #222a3d);
+  border-radius: 8px;
+  padding: 10px 14px;
+  font-size: 13px;
+  color: var(--text, #e6e9f0);
+}
+
+.sv-body-loading { padding: 8px 0 18px; }
+.sv-body {
+  background: var(--panel, #1a2030);
+  border: 1px solid var(--panel-2, #222a3d);
+  border-radius: 8px;
+  padding: 16px 18px;
+  font-size: 13px;
+  line-height: 1.7;
+  color: var(--text, #e6e9f0);
+  white-space: pre-wrap;
+  word-break: break-word;
+  margin-bottom: 18px;
+  max-height: 60vh;
+  overflow-y: auto;
+}
+.sv-snippet {
+  font-size: 13px;
+  line-height: 1.6;
+  color: var(--muted, #8b93a7);
+  margin-bottom: 18px;
+}
+.sv-detail-actions { display: flex; gap: 10px; flex-wrap: wrap; }
+
+/* ── List loading / empty states ───────────────────────────────────────── */
+.sv-skeleton-list {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+  padding: 12px 0;
+}
+.sv-skeleton-row {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  padding: 4px 8px;
+}
+
+/* ── Mobile: stack / overlay the reading pane ──────────────────────────── */
+@media (max-width: 767px) {
+  .sv-reading-pane {
+    position: fixed;
+    inset: 56px 0 0 0; /* below the topbar */
+    width: 100% !important;
+    max-width: 100vw;
+    min-width: 0;
+    border-left: none;
+    resize: none;
+    z-index: 40;
+  }
+  /* When the pane is open, hide the underlying list to avoid double-scroll. */
+  .sv-split.sv-split--open .sv-list-pane {
+    display: none;
+  }
+}

frontend/src/styles.css

@@ -1,18 +1,20 @@
 :root {
-  --bg: #0f1420;
+  /* v2 brand — warm charcoal + green (aligns with the token system in index.css).
-  --panel: #1a2030;
+     Legacy classes still read these; the v2 shell/components use the token system. */
-  --panel-2: #222a3d;
+  --bg: #1a1917;
-  --text: #e6e9f0;
+  --panel: #211f1d;
-  --muted: #8b93a7;
+  --panel-2: #2a2724;
-  --accent: #4f8cff;
+  --text: #f2efe9;
-  --danger: #eb5757;
+  --muted: #a8a29a;
-  --ok: #6fcf97;
+  --accent: #46ad27;
+  --danger: #d95a4a;
+  --ok: #5bbf4a;
 }
 
 * { box-sizing: border-box; }
 body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-serif; background: var(--bg); color: var(--text); }
 
-.topbar { display: flex; align-items: center; gap: 18px; padding: 12px 20px; background: var(--panel); border-bottom: 1px solid #2c3550; }
+.topbar { display: flex; align-items: center; gap: 18px; padding: 12px 20px; background: var(--panel); border-bottom: 1px solid #332f2b; }
 .brand { font-weight: 700; font-size: 18px; }
 .topbar nav a { color: var(--muted); text-decoration: none; margin-right: 14px; }
 .topbar nav a.active, .topbar nav a:hover { color: var(--text); }
@@ -21,13 +23,13 @@ body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-
 /* ── Search bar ── */
 .search-form { display: flex; align-items: center; gap: 0; flex: 1; max-width: 420px; }
 .search-input {
-  flex: 1; background: var(--panel-2); border: 1px solid #2c3550; border-right: none;
+  flex: 1; background: var(--panel-2); border: 1px solid #332f2b; border-right: none;
   color: var(--text); border-radius: 6px 0 0 6px; padding: 7px 10px; font-size: 13px;
   min-width: 0;
 }
 .search-input:focus { outline: none; border-color: var(--accent); }
 .search-btn {
-  background: var(--panel-2); border: 1px solid #2c3550; border-left: none;
+  background: var(--panel-2); border: 1px solid #332f2b; border-left: none;
   color: var(--muted); border-radius: 0 6px 6px 0; padding: 7px 10px; cursor: pointer; font-size: 13px;
 }
 .search-btn:hover { color: var(--text); border-color: var(--accent); }
@@ -38,7 +40,7 @@ body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-
 
 .sidebar {
   width: 216px; flex-shrink: 0;
-  background: var(--panel); border-right: 1px solid #2c3550;
+  background: var(--panel); border-right: 1px solid #332f2b;
   display: flex; flex-direction: column; overflow-y: auto;
   transition: width 0.2s ease;
 }
@@ -47,7 +49,7 @@ body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-
 .sidebar-header {
   display: flex; align-items: center; justify-content: space-between;
   padding: 12px 8px 10px;
-  border-bottom: 1px solid #2c3550;
+  border-bottom: 1px solid #332f2b;
   min-height: 42px; flex-shrink: 0;
 }
 .sidebar-brand { font-size: 12px; font-weight: 700; letter-spacing: 0.05em; text-transform: uppercase; color: var(--muted); padding-left: 6px; }
@@ -57,7 +59,7 @@ body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-
 /* Section headings */
 .section-head {
   display: flex; align-items: center; justify-content: space-between;
-  width: 100%; background: none; border: none; border-bottom: 1px solid #2c3550;
+  width: 100%; background: none; border: none; border-bottom: 1px solid #332f2b;
   padding: 7px 10px 7px 12px; cursor: pointer;
   color: var(--muted); font-size: 10px; font-weight: 700; letter-spacing: 0.07em; text-transform: uppercase;
 }
@@ -87,7 +89,7 @@ body { margin: 0; font-family: system-ui, -apple-system, Segoe UI, Roboto, sans-
   background: var(--panel-2); border-radius: 8px; padding: 1px 5px;
   flex-shrink: 0;
 }
-.folder-item--active .folder-badge { background: #2c3550; color: var(--accent); }
+.folder-item--active .folder-badge { background: #332f2b; color: var(--accent); }
 
 /* Favorite pin/unpin button */
 .fav-pin {
@@ -108,15 +110,13 @@ button:disabled { opacity: 0.5; cursor: default; }
 .toolbar { display: flex; align-items: center; gap: 10px; margin-bottom: 12px; flex-wrap: wrap; }
 .widget-toggles { display: flex; gap: 12px; flex-wrap: wrap; font-size: 12px; color: var(--muted); }
 
-.grid-item { background: var(--panel); border: 1px solid #2c3550; border-radius: 10px; overflow: hidden; }
+.grid-item { background: var(--panel); border: 1px solid #332f2b; border-radius: 10px; overflow: hidden; }
 .widget { padding: 14px; height: 100%; display: flex; flex-direction: column; }
 .widget h3 { margin: 0 0 10px; font-size: 14px; cursor: move; }
 .widget--link { cursor: pointer; }
 .widget--link:hover { border-color: var(--accent); }
 .mini-row--link { cursor: pointer; }
 .mini-row--link:hover td { color: var(--accent); }
-.chm-label--link { cursor: pointer; text-decoration: underline dotted; }
-.chm-label--link:hover { color: var(--accent); }
 .widget canvas { flex: 1; min-height: 0; }
 
 .stat { align-items: flex-start; justify-content: center; }
@@ -131,21 +131,25 @@ button:disabled { opacity: 0.5; cursor: default; }
 
 table.mini, table.grid { width: 100%; border-collapse: collapse; font-size: 13px; }
 table.mini td { padding: 3px 0; }
-table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; text-align: left; }
+table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #332f2b; text-align: left; }
 .num { text-align: right; }
 .muted { color: var(--muted); font-size: 12px; }
 
-.heatmap { display: flex; flex-direction: column; gap: 2px; }
+/* Emails by Category — ranked horizontal bars */
-.hm-row { display: flex; align-items: center; gap: 2px; }
+.cat-bars { display: flex; flex-direction: column; gap: 6px; overflow: auto; }
-.hm-day { width: 30px; font-size: 10px; color: var(--muted); }
+.cat-bar {
-.hm-cell { width: 10px; height: 10px; background: var(--accent); border-radius: 2px; }
+  display: grid; grid-template-columns: 92px 1fr 40px; gap: 8px; align-items: center;
-
+  width: 100%; padding: 3px 4px; margin: 0; border: none; background: transparent;
-/* Category heatmap */
+  border-radius: 6px; text-align: left; font: inherit; color: inherit;
-.cat-heatmap { display: flex; flex-direction: column; gap: 3px; overflow: auto; }
+}
-.chm-row { display: grid; grid-template-columns: 92px repeat(7, 1fr); gap: 3px; align-items: stretch; }
+.cat-bar--link { cursor: pointer; }
-.chm-head .chm-col { font-size: 10px; color: var(--muted); text-align: center; }
+.cat-bar--link:hover { background: rgba(255, 255, 255, 0.04); }
-.chm-label { font-size: 11px; color: var(--muted); overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.cat-bar:disabled { cursor: default; }
-.chm-cell { background: var(--accent); border-radius: 3px; min-height: 22px; display: flex; align-items: center; justify-content: center; font-size: 10px; color: #fff; }
+.cat-bar-label { font-size: 11px; color: var(--muted); overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
+.cat-bar-track { height: 14px; background: rgba(255, 255, 255, 0.06); border-radius: 4px; overflow: hidden; }
+.cat-bar-fill { display: block; height: 100%; background: var(--accent); border-radius: 4px; min-width: 2px; }
+.cat-bar-count { font-size: 11px; color: var(--muted); text-align: right; font-variant-numeric: tabular-nums; }
+.cat-bar--link:hover .cat-bar-label { color: var(--accent); }
 
 /* react-grid-layout resize handle — make it clearly visible on the dark theme */
 .react-resizable-handle {
@@ -172,7 +176,7 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 .sl-panel {
   display: flex;
   flex-direction: column;
-  border-right: 1px solid #2c3550;
+  border-right: 1px solid #332f2b;
   overflow: hidden;
 }
 .sl-search-wrap { padding: 10px 12px 6px; }
@@ -180,7 +184,7 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
   width: 100%;
   box-sizing: border-box;
   background: var(--panel-2);
-  border: 1px solid #2c3550;
+  border: 1px solid #332f2b;
   color: var(--text);
   border-radius: 6px;
   padding: 7px 10px;
@@ -191,7 +195,7 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 .sl-item {
   display: block; width: 100%; text-align: left;
   background: none; border: none; cursor: pointer;
-  padding: 10px 14px; border-bottom: 1px solid #1e2540;
+  padding: 10px 14px; border-bottom: 1px solid #26231f;
   color: var(--text);
 }
 .sl-item:hover { background: var(--panel-2); }
@@ -211,7 +215,7 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 }
 .sd-panel-header {
   padding: 16px 20px 12px;
-  border-bottom: 1px solid #2c3550;
+  border-bottom: 1px solid #332f2b;
   background: var(--panel);
   position: sticky; top: 0; z-index: 1;
 }
@@ -223,7 +227,7 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 .sd-detail { padding: 20px; max-width: 760px; }
 .sd-back {
   display: inline-flex; align-items: center; gap: 4px;
-  background: none; border: 1px solid #2c3550; border-radius: 6px;
+  background: none; border: 1px solid #332f2b; border-radius: 6px;
   color: var(--text); font-size: 13px; cursor: pointer;
   padding: 5px 12px; margin-bottom: 18px;
 }
@@ -232,9 +236,9 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 .sd-detail-subject { font-size: 18px; font-weight: 700; margin-bottom: 6px; }
 .sd-detail-meta { font-size: 12px; color: var(--muted); display: flex; flex-wrap: wrap; gap: 4px; align-items: center; }
 .sd-detail-sep { opacity: 0.4; }
-.sd-snippet { background: var(--panel); border: 1px solid #2c3550; border-radius: 8px; padding: 14px 16px; font-size: 13px; line-height: 1.6; color: var(--muted); white-space: pre-wrap; margin-bottom: 18px; }
+.sd-snippet { background: var(--panel); border: 1px solid #332f2b; border-radius: 8px; padding: 14px 16px; font-size: 13px; line-height: 1.6; color: var(--muted); white-space: pre-wrap; margin-bottom: 18px; }
 .sd-ai-summary { margin-bottom: 14px; }
-.sd-ai-summary-text { background: var(--panel-2); border: 1px solid #2c3550; border-radius: 8px; padding: 10px 14px; font-size: 13px; color: var(--text); }
+.sd-ai-summary-text { background: var(--panel-2); border: 1px solid #332f2b; border-radius: 8px; padding: 10px 14px; font-size: 13px; color: var(--text); }
 .sd-detail-actions { display: flex; gap: 10px; }
 .btn-sm {
   background: var(--accent); color: #fff; border: none; border-radius: 6px;
@@ -245,8 +249,8 @@ table.grid th, table.grid td { padding: 8px; border-bottom: 1px solid #2c3550; t
 .page { max-width: 900px; }
 .form-row { display: flex; gap: 10px; margin: 14px 0; }
 .form-row input { flex: 1; }
-input, select { background: var(--panel-2); border: 1px solid #2c3550; color: var(--text); border-radius: 6px; padding: 8px; }
+input, select { background: var(--panel-2); border: 1px solid #332f2b; color: var(--text); border-radius: 6px; padding: 8px; }
-.card { background: var(--panel); border: 1px solid #2c3550; border-radius: 10px; padding: 14px; margin-top: 12px; }
+.card { background: var(--panel); border: 1px solid #332f2b; border-radius: 10px; padding: 14px; margin-top: 12px; }
 .card.success { border-color: var(--ok); }
 .card ul { font-size: 13px; color: var(--muted); }
 
@@ -259,14 +263,14 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 
 /* ── Shared buttons ── */
 .brand-link { text-decoration: none; color: var(--text); display: inline-flex; }
-.ghost { background: transparent; border: 1px solid #36405c; color: var(--text); }
+.ghost { background: transparent; border: 1px solid #3d3833; color: var(--text); }
 .ghost:hover { border-color: var(--accent); }
 .cta {
   background: var(--accent); color: #fff; border: none; border-radius: 8px;
   padding: 12px 22px; font-size: 16px; font-weight: 700; cursor: pointer;
   text-decoration: none; display: inline-block;
 }
-.cta:hover { background: #5d97ff; }
+.cta:hover { background: #57c231; }
 
 /* ── Landing page ── */
 .landing { max-width: 1080px; margin: 0 auto; padding: 0 20px 60px; }
@@ -277,13 +281,13 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .hero-cta { margin: 8px 0; }
 .fineprint { color: var(--muted); font-size: 12px; margin-top: 14px; }
 .features { display: grid; grid-template-columns: repeat(auto-fit, minmax(280px, 1fr)); gap: 16px; margin-top: 24px; }
-.feature { background: var(--panel); border: 1px solid #2c3550; border-radius: 12px; padding: 20px; }
+.feature { background: var(--panel); border: 1px solid #332f2b; border-radius: 12px; padding: 20px; }
 .feature-icon { font-size: 26px; }
 .feature h3 { margin: 10px 0 6px; font-size: 17px; }
 .feature p { color: var(--muted); font-size: 14px; line-height: 1.5; margin: 0; }
 .closing { text-align: center; margin: 56px 0 20px; }
 .closing h2 { font-size: 28px; margin-bottom: 18px; }
-.landing-footer { display: flex; align-items: center; gap: 12px; justify-content: center; margin-top: 40px; padding-top: 20px; border-top: 1px solid #2c3550; }
+.landing-footer { display: flex; align-items: center; gap: 12px; justify-content: center; margin-top: 40px; padding-top: 20px; border-top: 1px solid #332f2b; }
 
 /* ── Folder view ── */
 .folder-view { max-width: 960px; }
@@ -294,32 +298,57 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .fv-error { color: var(--danger); font-size: 14px; padding: 12px 0; }
 
 .email-list { width: 100%; border-collapse: collapse; font-size: 13px; }
-.email-row { border-bottom: 1px solid #2c3550; cursor: pointer; }
+.email-row {
+  height: 44px;
+  border-bottom: 1px solid #332f2b;
+  cursor: pointer;
+  transition: background 0.1s ease;
+}
+.email-row > td { padding-top: 0; padding-bottom: 0; vertical-align: middle; }
+/* Single, consistent hover state for the whole row. */
 .email-row:hover { background: var(--panel); }
-.email-row--unread .el-sender,
+/* Unread: prominent subject, keep sender readable but not shouty. */
 .email-row--unread .el-subj-text { font-weight: 700; color: var(--text); }
+.email-row--unread .el-sender { font-weight: 600; color: var(--text); }
 
-.el-unread { width: 14px; padding: 10px 4px 10px 0; }
+.el-select { width: 34px; padding: 0 4px 0 10px; text-align: center; }
+.el-select > * { vertical-align: middle; }
+
+.el-unread { width: 14px; padding: 0 4px 0 0; text-align: center; }
 .unread-dot { display: inline-block; width: 7px; height: 7px; border-radius: 50%; background: var(--accent); }
-.el-sender { width: 180px; padding: 10px 12px 10px 4px; white-space: nowrap; overflow: hidden; text-overflow: ellipsis; color: var(--muted); }
+
-.el-subject { padding: 10px 8px; overflow: hidden; }
+/* Sender: secondary in the hierarchy — muted by default. */
+.el-sender {
+  width: 180px; padding: 0 12px 0 4px;
+  white-space: nowrap; overflow: hidden; text-overflow: ellipsis;
+  color: var(--muted); font-size: 12.5px;
+}
+
+/* Subject + snippet share one line: subject prominent, snippet muted. */
+.el-subject { padding: 0 8px; overflow: hidden; max-width: 0; white-space: nowrap; text-overflow: ellipsis; }
 .el-subj-text { color: var(--text); }
-.el-snippet { color: var(--muted); }
+.el-snippet {
-.el-meta { width: 80px; padding: 10px 8px; text-align: right; white-space: nowrap; }
+  color: var(--muted); font-size: 12.5px;
+}
+.el-snippet::before { content: '—'; margin: 0 6px; opacity: 0.55; }
+
+.el-meta { width: 80px; padding: 0 8px; text-align: right; white-space: nowrap; }
 .el-attach { margin-right: 4px; font-size: 12px; }
 .el-size { font-size: 11px; color: var(--muted); }
-.el-date { width: 70px; padding: 10px 0 10px 8px; text-align: right; color: var(--muted); white-space: nowrap; font-size: 12px; }
+.el-date { width: 70px; padding: 0 0 0 8px; text-align: right; color: var(--muted); white-space: nowrap; font-size: 12px; }
 
-.el-actions { width: 80px; padding: 0 6px; text-align: right; white-space: nowrap; }
+.el-actions { width: 96px; padding: 0 6px; text-align: right; white-space: nowrap; }
 .action-btn {
-  background: none; border: none; padding: 3px 4px; cursor: pointer;
+  background: none; border: none; padding: 4px 5px; cursor: pointer;
-  font-size: 13px; opacity: 0; transition: opacity 0.1s, color 0.1s;
+  font-size: 13px; line-height: 1; opacity: 0;
+  transition: opacity 0.1s ease, color 0.1s ease, background 0.1s ease;
   border-radius: 4px; color: var(--muted);
 }
-.action-btn:hover { background: var(--panel-2); opacity: 1 !important; }
+.action-btn:hover { background: var(--panel-2); color: var(--text); opacity: 1 !important; }
 .action-btn--active { opacity: 1 !important; }
 .action-btn--danger:hover { color: var(--danger); }
-.email-row:hover .action-btn { opacity: 0.6; }
+.email-row:hover .action-btn { opacity: 0.65; }
+.action-btn:focus-visible { opacity: 1 !important; outline: 2px solid var(--accent); outline-offset: 1px; }
 .email-row--acting { opacity: 0.6; pointer-events: none; }
 .action-btn--unsub { font-size: 11px; }
 .action-btn--done { opacity: 1 !important; color: var(--ok); }
@@ -337,7 +366,7 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .splash-card { text-align: center; max-width: 440px; padding: 32px; }
 .splash-card h2 { margin: 16px 0 6px; }
 .progress-track { height: 10px; background: var(--panel-2); border-radius: 6px; overflow: hidden; margin: 20px 0 10px; }
-.progress-fill { height: 100%; background: linear-gradient(90deg, #4f8cff, #6fcf97); border-radius: 6px; transition: width 0.4s ease; }
+.progress-fill { height: 100%; background: linear-gradient(90deg, #46ad27, #6fcf97); border-radius: 6px; transition: width 0.4s ease; }
 .progress-fill[data-indeterminate="true"] { animation: indet 1.2s ease-in-out infinite; }
 @keyframes indet { 0% { margin-left: -40%; } 100% { margin-left: 100%; } }
 .progress-label { color: var(--muted); font-size: 13px; }
@@ -348,7 +377,7 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .sync-label--err { color: var(--danger); }
 .sync-spinner {
   width: 10px; height: 10px; border-radius: 50%;
-  border: 2px solid #36405c; border-top-color: var(--accent);
+  border: 2px solid #3d3833; border-top-color: var(--accent);
   animation: sync-spin 0.7s linear infinite;
 }
 @keyframes sync-spin { to { transform: rotate(360deg); } }
@@ -358,7 +387,7 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 /* ── Email detail body (Senders page) ──────────────────────────────────── */
 .sd-body-loading { padding: 8px 0 18px; }
 .sd-body {
-  background: var(--panel); border: 1px solid #2c3550; border-radius: 8px;
+  background: var(--panel); border: 1px solid #332f2b; border-radius: 8px;
   padding: 16px 18px; font-size: 13px; line-height: 1.7; color: var(--text);
   white-space: pre-wrap; word-break: break-word; margin-bottom: 18px;
   max-height: 60vh; overflow-y: auto;
@@ -369,7 +398,7 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .unsub-toast { color: var(--ok); font-size: 13px; margin-left: 6px; }
 .unsub-tabs { display: flex; gap: 6px; margin: 16px 0 10px; }
 .unsub-tab {
-  background: var(--panel); border: 1px solid #2c3550; color: var(--muted);
+  background: var(--panel); border: 1px solid #332f2b; color: var(--muted);
   border-radius: 6px; padding: 6px 12px; font-size: 13px; cursor: pointer;
   display: flex; align-items: center; gap: 6px;
 }
@@ -377,7 +406,7 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 .unsub-tab--active { color: var(--text); border-color: var(--accent); background: var(--panel-2); }
 .unsub-tab-count { font-size: 11px; color: var(--muted); background: var(--bg); border-radius: 8px; padding: 1px 6px; }
 .unsub-grid { width: 100%; }
-.unsub-badge { font-size: 11px; padding: 2px 8px; border-radius: 10px; border: 1px solid #36405c; }
+.unsub-badge { font-size: 11px; padding: 2px 8px; border-radius: 10px; border: 1px solid #3d3833; }
 .unsub-badge--detected { color: var(--muted); }
 .unsub-badge--queued { color: #f2c94c; border-color: #f2c94c66; }
 .unsub-badge--progress { color: var(--accent); border-color: var(--accent); }
@@ -392,27 +421,26 @@ input, select { background: var(--panel-2); border: 1px solid #2c3550; color: va
 /* ── Bulk selection toolbar (folder/search/sender email lists) ──────────── */
 .bulk-toolbar {
   display: flex; align-items: center; gap: 10px;
-  background: var(--panel-2); border: 1px solid #2c3550; border-radius: 8px;
+  background: var(--panel-2); border: 1px solid #332f2b; border-radius: 8px;
   padding: 8px 14px; margin-bottom: 10px; font-size: 13px;
 }
 .bulk-toolbar .muted { font-size: 12px; }
 .bulk-toolbar-spacer { flex: 1; }
 .bulk-btn {
-  background: var(--panel); border: 1px solid #2c3550; color: var(--text);
+  background: var(--panel); border: 1px solid #332f2b; color: var(--text);
   border-radius: 6px; padding: 5px 12px; font-size: 12px; cursor: pointer;
 }
 .bulk-btn:hover { border-color: var(--accent); }
 .bulk-btn--danger:hover { border-color: var(--danger); color: var(--danger); }
-.el-select { width: 28px; text-align: center; }
 
 /* ── Keyboard shortcut help ──────────────────────────────────────────────── */
-.kbd-hint { position: fixed; bottom: 14px; right: 14px; font-size: 11px; color: var(--muted); background: var(--panel); border: 1px solid #2c3550; border-radius: 6px; padding: 4px 10px; opacity: 0.7; }
+.kbd-hint { position: fixed; bottom: 14px; right: 14px; font-size: 11px; color: var(--muted); background: var(--panel); border: 1px solid #332f2b; border-radius: 6px; padding: 4px 10px; opacity: 0.7; }
-.kbd-hint kbd { background: var(--panel-2); border: 1px solid #36405c; border-radius: 3px; padding: 0 4px; font-family: inherit; }
+.kbd-hint kbd { background: var(--panel-2); border: 1px solid #3d3833; border-radius: 3px; padding: 0 4px; font-family: inherit; }
 
 /* ── Saved searches ─────────────────────────────────────────────────────── */
 .saved-search-row { display: flex; align-items: center; gap: 8px; }
 .saved-search-save-btn {
-  background: none; border: 1px solid #2c3550; color: var(--muted);
+  background: none; border: 1px solid #332f2b; color: var(--muted);
   border-radius: 6px; padding: 6px 10px; font-size: 12px; cursor: pointer;
 }
 .saved-search-save-btn:hover { color: var(--text); border-color: var(--accent); }

frontend/tailwind.config.js

@@ -31,6 +31,32 @@ export default {
           DEFAULT: 'hsl(var(--danger) / <alpha-value>)',
           foreground: 'hsl(var(--danger-foreground) / <alpha-value>)',
         },
+        // Static brand ramp (from #3ba31f) for utility use (bg-green-500, etc.).
+        // The theme-aware accent above (`primary`) is what most UI should use.
+        green: {
+          50: '#f1f9ec',
+          100: '#dcf0d0',
+          200: '#bde3ab',
+          300: '#93d07d',
+          400: '#63b84a',
+          500: '#3ba31f',
+          600: '#2f8419',
+          700: '#266a15',
+          800: '#1e5312',
+          900: '#163a0e',
+        },
+      },
+      fontFamily: {
+        sans: [
+          'Inter Variable',
+          'Inter',
+          'ui-sans-serif',
+          'system-ui',
+          '-apple-system',
+          'Segoe UI',
+          'Roboto',
+          'sans-serif',
+        ],
       },
       borderRadius: {
         lg: 'var(--radius)',

renovate.json

@@ -0,0 +1,27 @@
+{
+  "$schema": "https://docs.renovatebot.com/renovate-schema.json",
+  "extends": ["config:recommended"],
+  "timezone": "Europe/Berlin",
+  "schedule": ["before 6am on monday"],
+  "labels": ["dependencies"],
+  "prConcurrentLimit": 5,
+  "commitMessagePrefix": "chore(deps):",
+  "packageRules": [
+    {
+      "description": "Group safe minor+patch updates into one weekly PR per ecosystem",
+      "matchUpdateTypes": ["minor", "patch"],
+      "groupName": "{{manager}} minor & patch"
+    },
+    {
+      "description": "Major updates stay individual PRs for careful review",
+      "matchUpdateTypes": ["major"],
+      "dependencyDashboardApproval": true
+    }
+  ],
+  "vulnerabilityAlerts": {
+    "enabled": true,
+    "labels": ["security"],
+    "schedule": ["at any time"]
+  },
+  "ignorePaths": ["**/node_modules/**", "**/bin/**", "**/obj/**"]
+}

scripts/git-hooks/pre-commit

@@ -0,0 +1,43 @@
+#!/usr/bin/env sh
+# InboxIntel pre-commit hook — fast checks only (keep it under a few seconds).
+# Heavier build/test verification runs in pre-push. Bypass in an emergency with:
+#   git commit --no-verify
+set -eu
+
+echo "[pre-commit] running fast checks..."
+
+# 1) Block obvious secrets from being committed. Scans only staged, added lines.
+STAGED=$(git diff --cached --name-only --diff-filter=ACM)
+if [ -n "$STAGED" ]; then
+  # Never allow a real .env (only *.example templates are tracked).
+  echo "$STAGED" | grep -E '(^|/)\.env($|\.)' | grep -v '\.example$' >/tmp/ii_env_hits 2>/dev/null || true
+  if [ -s /tmp/ii_env_hits ]; then
+    echo "[pre-commit] BLOCKED: attempting to commit an env/secret file:"
+    cat /tmp/ii_env_hits
+    echo "  -> add it to .gitignore or commit .env.example instead."
+    exit 1
+  fi
+
+  # Heuristic secret scan on added lines (private keys, obvious credential assigns).
+  if git diff --cached --unified=0 -- $STAGED \
+      | grep -E '^\+' \
+      | grep -Ei 'BEGIN (RSA|OPENSSH|EC|PGP) PRIVATE KEY|(client_secret|password|api[_-]?key|secret)\s*[:=]\s*["'"'"']?[A-Za-z0-9/_+=-]{16,}' \
+      | grep -Evi 'change-me|your-|example|placeholder|\$\{' >/tmp/ii_secret_hits 2>/dev/null; then
+    echo "[pre-commit] BLOCKED: possible hard-coded secret in staged changes:"
+    cat /tmp/ii_secret_hits
+    echo "  -> use deploy/.env / user-secrets. Override with 'git commit --no-verify' if this is a false positive."
+    exit 1
+  fi
+fi
+
+# 2) Format .NET code (only if the tool + staged .cs files are present). Verify-only,
+#    non-mutating, so it never rewrites files out from under your staged diff.
+if echo "$STAGED" | grep -q '\.cs$'; then
+  if command -v dotnet >/dev/null 2>&1 && dotnet format --help >/dev/null 2>&1; then
+    echo "[pre-commit] dotnet format --verify-no-changes"
+    dotnet format InboxIntel.sln --verify-no-changes --verbosity quiet \
+      || { echo "  -> run 'dotnet format InboxIntel.sln' and re-stage."; exit 1; }
+  fi
+fi
+
+echo "[pre-commit] OK"

scripts/git-hooks/pre-push

@@ -0,0 +1,21 @@
+#!/usr/bin/env sh
+# InboxIntel pre-push hook — build + test gate. Mirrors what Gitea CI runs so you
+# catch failures before they reach the server. Bypass with: git push --no-verify
+set -eu
+
+echo "[pre-push] building solution (Release)..."
+dotnet build InboxIntel.sln -c Release --nologo \
+  || { echo "[pre-push] BLOCKED: backend build failed."; exit 1; }
+
+echo "[pre-push] running tests..."
+dotnet test InboxIntel.sln -c Release --no-build --nologo \
+  || { echo "[pre-push] BLOCKED: tests failed."; exit 1; }
+
+# Frontend build (only if the app is present and npm is installed).
+if [ -f frontend/package.json ] && command -v npm >/dev/null 2>&1; then
+  echo "[pre-push] frontend build..."
+  ( cd frontend && npm run build --silent ) \
+    || { echo "[pre-push] BLOCKED: frontend build failed."; exit 1; }
+fi
+
+echo "[pre-push] OK — safe to push."

scripts/install-hooks.ps1

@@ -0,0 +1,24 @@
+<#
+.SYNOPSIS
+  Point git at the version-controlled hooks in scripts/git-hooks.
+.DESCRIPTION
+  Uses `git config core.hooksPath` so the hooks live in the repo (reviewable,
+  shared, updatable) instead of the un-tracked .git/hooks directory. Run once
+  per clone. Git for Windows ships the bash needed to execute the POSIX hooks.
+.EXAMPLE
+  ./scripts/install-hooks.ps1
+#>
+$ErrorActionPreference = 'Stop'
+$root = Split-Path -Parent $PSScriptRoot
+Push-Location $root
+try {
+  git config core.hooksPath scripts/git-hooks
+  # Best-effort exec bit (matters on Linux/WSL; harmless on Windows). Only applies
+  # once the hooks are tracked; ignored on a first run before they're committed.
+  foreach ($h in 'pre-commit','pre-push') {
+    try { git update-index --chmod=+x "scripts/git-hooks/$h" 2>$null } catch {}
+  }
+  Write-Host "Installed git hooks -> scripts/git-hooks (core.hooksPath set)." -ForegroundColor Green
+  Write-Host "Bypass in an emergency with --no-verify." -ForegroundColor DarkGray
+}
+finally { Pop-Location }

src/InboxIntel.Api/Controllers/AiController.cs

@@ -7,6 +7,7 @@ namespace InboxIntel.Api.Controllers;
 /// AI endpoints are read-only / advisory. They never trigger destructive
 /// actions - suggestions are returned for the user to act on via /cleanup.
 /// </summary>
+[Microsoft.AspNetCore.RateLimiting.EnableRateLimiting("expensive")]  // AUDIT H-2: LLM calls are the most expensive path
 public class AiController : ApiControllerBase
 {
     private readonly IAiService _ai;

src/InboxIntel.Api/Controllers/AnalyticsController.cs

@@ -24,9 +24,6 @@ public class AnalyticsController : ApiControllerBase
     public async Task<IActionResult> Volume([FromQuery] int days = 90, CancellationToken ct = default)
         => Ok(await _analytics.GetVolumeOverTimeAsync(UserId, Math.Clamp(days, 1, 3660), ct));
 
-    [HttpGet("heatmap")]
-    public async Task<IActionResult> Heatmap(CancellationToken ct) => Ok(await _analytics.GetHeatmapAsync(UserId, ct));
-
     [HttpGet("category-heatmap")]
     public async Task<IActionResult> CategoryHeatmap(CancellationToken ct) => Ok(await _analytics.GetCategoryHeatmapAsync(UserId, ct));
 

src/InboxIntel.Api/Controllers/AuthController.cs

@@ -5,10 +5,12 @@ using Microsoft.AspNetCore.Authentication.Cookies;
 using Microsoft.AspNetCore.Authentication.Google;
 using Microsoft.AspNetCore.Authorization;
 using Microsoft.AspNetCore.Mvc;
+using Microsoft.AspNetCore.RateLimiting;
 
 namespace InboxIntel.Api.Controllers;
 
 [ApiController]
+[EnableRateLimiting("auth")]  // AUDIT H-2: throttle login/challenge attempts per IP
 [ApiVersion("1.0")]
 [Route("api/v{version:apiVersion}/[controller]")]
 public class AuthController : ControllerBase

src/InboxIntel.Api/Controllers/EmailController.cs

@@ -72,6 +72,22 @@ public class EmailController : ApiControllerBase
     [HttpPost("{id:guid}/untrash")]
     public Task<IActionResult> Untrash(Guid id, CancellationToken ct) => Act(id, CleanupActionType.Archive, ct);
 
+    /// <summary>Toggle the local-only Read Later marker (no Gmail side effect).</summary>
+    [HttpPost("{id:guid}/readlater")]
+    public Task<IActionResult> ReadLater(Guid id, CancellationToken ct) => SetReadLater(id, true, ct);
+
+    [HttpPost("{id:guid}/unreadlater")]
+    public Task<IActionResult> UnreadLater(Guid id, CancellationToken ct) => SetReadLater(id, false, ct);
+
+    private async Task<IActionResult> SetReadLater(Guid id, bool value, CancellationToken ct)
+    {
+        var email = await _db.Emails.FirstOrDefaultAsync(e => e.Id == id && e.UserId == UserId, ct);
+        if (email is null) return NotFound();
+        email.IsReadLater = value;
+        await _db.SaveChangesAsync(ct);
+        return Ok();
+    }
+
     /// <summary>
     /// Inline unsubscribe. Detects the unsubscribe mechanism for the email's sender,
     /// then executes it (HTTP one-click or HTTP link). mailto targets cannot be sent

src/InboxIntel.Api/Controllers/ExportController.cs

@@ -3,6 +3,7 @@ using Microsoft.AspNetCore.Mvc;
 
 namespace InboxIntel.Api.Controllers;
 
+[Microsoft.AspNetCore.RateLimiting.EnableRateLimiting("expensive")]  // AUDIT H-2: PDF/CSV generation is costly
 public class ExportController : ApiControllerBase
 {
     private readonly IExportService _export;

src/InboxIntel.Api/Controllers/UnsubscribeController.cs

@@ -4,6 +4,7 @@ using Microsoft.AspNetCore.Mvc;
 
 namespace InboxIntel.Api.Controllers;
 
+[Microsoft.AspNetCore.RateLimiting.EnableRateLimiting("expensive")]  // AUDIT H-2: triggers server-side outbound HTTP
 public class UnsubscribeController : ApiControllerBase
 {
     private readonly IUnsubscribeService _unsub;

src/InboxIntel.Api/Controllers/WidgetLayoutController.cs

@@ -42,8 +42,15 @@ public class WidgetLayoutController : ApiControllerBase
             {
                 _db.WidgetLayouts.Add(new WidgetLayout
                 {
-                    UserId = UserId, WidgetKey = dto.WidgetKey, X = dto.X, Y = dto.Y, W = dto.W, H = dto.H,
+                    UserId = UserId,
-                    Visible = dto.Visible, SortOrder = dto.SortOrder, SettingsJson = dto.SettingsJson
+                    WidgetKey = dto.WidgetKey,
+                    X = dto.X,
+                    Y = dto.Y,
+                    W = dto.W,
+                    H = dto.H,
+                    Visible = dto.Visible,
+                    SortOrder = dto.SortOrder,
+                    SettingsJson = dto.SettingsJson
                 });
             }
         }

src/InboxIntel.Api/Dockerfile

@@ -1,5 +1,5 @@
 # Multi-stage build for the ASP.NET Core API.
-FROM mcr.microsoft.com/dotnet/sdk:8.0 AS build
+FROM mcr.microsoft.com/dotnet/sdk:10.0 AS build
 WORKDIR /src
 
 # Copy solution + project files first for layer-cached restore.
@@ -13,8 +13,16 @@ RUN dotnet restore src/InboxIntel.Api/InboxIntel.Api.csproj
 COPY src/ src/
 RUN dotnet publish src/InboxIntel.Api/InboxIntel.Api.csproj -c Release -o /app/publish /p:UseAppHost=false
 
-FROM mcr.microsoft.com/dotnet/aspnet:8.0 AS runtime
+FROM mcr.microsoft.com/dotnet/aspnet:10.0 AS runtime
 WORKDIR /app
+
+# The slim aspnet:10.0 image dropped libgssapi_krb5, which Npgsql tries to load during
+# connection negotiation ("Cannot load library libgssapi_krb5.so.2"). Harmless for password
+# auth but noisy and a latent failure on some paths — install the Kerberos runtime lib.
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends libgssapi-krb5-2 \
+    && rm -rf /var/lib/apt/lists/*
+
 COPY --from=build /app/publish .
 
 # V-12: run as the non-root 'app' user shipped in the .NET 8 images. Pre-create the

src/InboxIntel.Api/InboxIntel.Api.csproj

@@ -5,16 +5,21 @@
     <UserSecretsId>210c6d96-c7e4-4ee9-8982-8b91424979b8</UserSecretsId>
   </PropertyGroup>
   <ItemGroup>
-    <PackageReference Include="Microsoft.AspNetCore.Authentication.Google" Version="8.0.7" />
+    <PackageReference Include="Microsoft.AspNetCore.Authentication.Google" Version="10.0.9" />
-    <PackageReference Include="Microsoft.AspNetCore.Authentication.JwtBearer" Version="8.0.7" />
+    <PackageReference Include="Microsoft.AspNetCore.Authentication.JwtBearer" Version="10.0.9" />
     <!-- Required on the startup project for `dotnet ef migrations` to work. -->
-    <PackageReference Include="Microsoft.EntityFrameworkCore.Design" Version="8.0.4">
+    <PackageReference Include="Microsoft.EntityFrameworkCore.Design" Version="10.0.9">
       <IncludeAssets>runtime; build; native; contentfiles; analyzers; buildtransitive</IncludeAssets>
       <PrivateAssets>all</PrivateAssets>
     </PackageReference>
     <PackageReference Include="Asp.Versioning.Mvc" Version="8.1.0" />
     <PackageReference Include="Asp.Versioning.Mvc.ApiExplorer" Version="8.1.0" />
     <PackageReference Include="FluentValidation.AspNetCore" Version="11.3.0" />
+    <PackageReference Include="Npgsql.OpenTelemetry" Version="10.0.3" />
+    <PackageReference Include="OpenTelemetry.Exporter.OpenTelemetryProtocol" Version="1.16.0" />
+    <PackageReference Include="OpenTelemetry.Extensions.Hosting" Version="1.16.0" />
+    <PackageReference Include="OpenTelemetry.Instrumentation.AspNetCore" Version="1.16.0" />
+    <PackageReference Include="OpenTelemetry.Instrumentation.Http" Version="1.16.0" />
     <PackageReference Include="Serilog.AspNetCore" Version="8.0.1" />
     <PackageReference Include="Serilog.Sinks.Console" Version="5.0.1" />
     <PackageReference Include="Swashbuckle.AspNetCore" Version="6.6.2" />

src/InboxIntel.Api/Program.cs

@@ -1,5 +1,7 @@
 using System.Security.Claims;
+using System.Threading.RateLimiting;
 using Asp.Versioning;
+using FluentValidation.AspNetCore;
 using InboxIntel.Api.Auth;
 using InboxIntel.Application;
 using InboxIntel.Application.Abstractions;
@@ -8,10 +10,16 @@ using InboxIntel.Infrastructure.Configuration;
 using InboxIntel.Infrastructure.Persistence;
 using Microsoft.AspNetCore.Authentication.Cookies;
 using Microsoft.AspNetCore.Authentication.Google;
+using Microsoft.AspNetCore.Authentication;
 using Microsoft.AspNetCore.DataProtection;
 using Microsoft.AspNetCore.HttpOverrides;
+using Microsoft.AspNetCore.RateLimiting;
 using Microsoft.EntityFrameworkCore;
 using Microsoft.Extensions.Options;
+using Npgsql;
+using OpenTelemetry.Metrics;
+using OpenTelemetry.Resources;
+using OpenTelemetry.Trace;
 using Serilog;
 
 var builder = WebApplication.CreateBuilder(args);
@@ -23,9 +31,20 @@ builder.Host.UseSerilog((ctx, cfg) => cfg
     .WriteTo.Console());
 
 // Persist Data Protection keys so encrypted refresh tokens survive restarts.
-builder.Services.AddDataProtection()
+// AUDIT M-3: optionally encrypt the Data Protection key ring with an X.509 certificate so
+// the keys are not readable in plaintext from the /keys volume (which would otherwise let
+// anyone with volume access decrypt all stored refresh tokens). Configure
+// DataProtection:CertificatePath (+ CertificatePassword) to enable; without it, keys are
+// persisted unprotected and a startup warning documents the residual risk.
+var dp = builder.Services.AddDataProtection()
     .PersistKeysToFileSystem(new DirectoryInfo(builder.Configuration["DataProtection:KeyPath"] ?? "/keys"))
     .SetApplicationName("InboxIntel");
+var dpCertPath = builder.Configuration["DataProtection:CertificatePath"];
+if (!string.IsNullOrWhiteSpace(dpCertPath))
+{
+    dp.ProtectKeysWithCertificate(System.Security.Cryptography.X509Certificates.X509CertificateLoader
+        .LoadPkcs12FromFile(dpCertPath, builder.Configuration["DataProtection:CertificatePassword"]));
+}
 
 builder.Services.AddApplication();
 builder.Services.AddInfrastructure(builder.Configuration);
@@ -57,6 +76,24 @@ builder.Services.AddAuthentication(options =>
         options.Cookie.Name = "inboxintel.session";
         options.ExpireTimeSpan = TimeSpan.FromDays(7);
         options.SlidingExpiration = true;
+        // AUDIT M-1: sliding expiration alone lets a stolen cookie renew forever. Stamp an
+        // absolute start at sign-in and reject principals older than the configured cap,
+        // forcing a full re-login. (Pre-existing sessions without the stamp are rejected
+        // once — a single re-login, then they carry the stamp.)
+        var absoluteDays = builder.Configuration.GetValue("Auth:AbsoluteSessionDays", 30);
+        options.Events.OnSigningIn = ctx =>
+        {
+            ctx.Properties.SetString("abs-start", DateTimeOffset.UtcNow.ToString("O"));
+            return Task.CompletedTask;
+        };
+        options.Events.OnValidatePrincipal = async ctx =>
+        {
+            if (SessionLifetime.IsExpired(ctx.Properties.GetString("abs-start"), DateTimeOffset.UtcNow, TimeSpan.FromDays(absoluteDays)))
+            {
+                ctx.RejectPrincipal();
+                await ctx.HttpContext.SignOutAsync(CookieAuthenticationDefaults.AuthenticationScheme);
+            }
+        };
         // API-style behaviour: return status codes rather than redirecting to a login page.
         options.Events.OnRedirectToLogin = ctx =>
         {
@@ -89,6 +126,28 @@ builder.Services.AddAuthentication(options =>
 
 builder.Services.AddAuthorization();
 
+// RECOMMENDATIONS #5: OpenTelemetry traces + metrics (ASP.NET, outbound HTTP, Npgsql).
+// The OTLP exporter only activates when Otel:Endpoint (or the standard
+// OTEL_EXPORTER_OTLP_ENDPOINT env var) is configured — zero overhead otherwise.
+// Logs stay on Serilog. Pair with the compose "observability" profile (grafana/otel-lgtm).
+var otlpEndpoint = builder.Configuration["Otel:Endpoint"]
+    ?? Environment.GetEnvironmentVariable("OTEL_EXPORTER_OTLP_ENDPOINT");
+if (!string.IsNullOrWhiteSpace(otlpEndpoint))
+{
+    builder.Services.AddOpenTelemetry()
+        .ConfigureResource(r => r.AddService("inboxintel-api"))
+        .WithTracing(t => t
+            .AddAspNetCoreInstrumentation()
+            .AddHttpClientInstrumentation()
+            .AddNpgsql()
+            .AddOtlpExporter(o => o.Endpoint = new Uri(otlpEndpoint)))
+        .WithMetrics(m => m
+            .AddAspNetCoreInstrumentation()
+            .AddHttpClientInstrumentation()
+            .AddNpgsqlInstrumentation()
+            .AddOtlpExporter(o => o.Endpoint = new Uri(otlpEndpoint)));
+}
+
 builder.Services.AddApiVersioning(o =>
 {
     o.DefaultApiVersion = new ApiVersion(1, 0);
@@ -98,6 +157,35 @@ builder.Services.AddApiVersioning(o =>
 }).AddApiExplorer(o => { o.GroupNameFormat = "'v'VVV"; o.SubstituteApiVersionInUrl = true; });
 
 builder.Services.AddControllers();
+// AUDIT H-1: the validators in InboxIntel.Application/Validation were registered but never
+// executed (FluentValidation 11.x needs explicit auto-validation). This wires them into
+// model binding so invalid DTOs 400 at the boundary instead of reaching services.
+builder.Services.AddFluentValidationAutoValidation();
+
+// AUDIT H-2: rate limiting. Global per-user (or per-IP when anonymous) window, plus stricter
+// named policies for auth and expensive endpoints (export/unsubscribe/AI). Limits are
+// config-driven so tests and deployments can tune them.
+var rl = builder.Configuration.GetSection("RateLimiting");
+int Limit(string key, int def) => rl.GetValue(key, def);
+var rlWindow = TimeSpan.FromSeconds(Limit("WindowSeconds", 60));
+static string Partition(HttpContext ctx) =>
+    ctx.User.Identity?.IsAuthenticated == true
+        ? ctx.User.FindFirstValue("inboxintel:uid") ?? "auth-unknown"
+        : ctx.Connection.RemoteIpAddress?.ToString() ?? "anon";
+builder.Services.AddRateLimiter(o =>
+{
+    o.RejectionStatusCode = StatusCodes.Status429TooManyRequests;
+    o.GlobalLimiter = PartitionedRateLimiter.Create<HttpContext, string>(ctx =>
+        RateLimitPartition.GetFixedWindowLimiter(Partition(ctx), _ =>
+            new FixedWindowRateLimiterOptions { PermitLimit = Limit("GlobalPermitLimit", 300), Window = rlWindow, QueueLimit = 0 }));
+    o.AddPolicy("auth", ctx =>
+        RateLimitPartition.GetFixedWindowLimiter(Partition(ctx), _ =>
+            new FixedWindowRateLimiterOptions { PermitLimit = Limit("AuthPermitLimit", 10), Window = rlWindow, QueueLimit = 0 }));
+    o.AddPolicy("expensive", ctx =>
+        RateLimitPartition.GetFixedWindowLimiter(Partition(ctx), _ =>
+            new FixedWindowRateLimiterOptions { PermitLimit = Limit("ExpensivePermitLimit", 20), Window = rlWindow, QueueLimit = 0 }));
+});
+
 builder.Services.AddEndpointsApiExplorer();
 builder.Services.AddSwaggerGen();
 // V-06: RFC7807 ProblemDetails so the global exception handler returns a safe,
@@ -115,7 +203,18 @@ using (var scope = app.Services.CreateScope())
 {
     var db = scope.ServiceProvider.GetRequiredService<AppDbContext>();
     if (db.Database.IsRelational() && app.Configuration.GetValue("Database:AutoMigrate", true))
+    {
+        // AUDIT M-4: the shipped appsettings no longer carries a guessable default DB
+        // password. Fail fast with a clear message rather than connecting with weak or
+        // missing credentials (compose/staging/prod inject the full connection string).
+        var connStr = app.Configuration.GetConnectionString("Postgres") ?? string.Empty;
+        var csb = new Npgsql.NpgsqlConnectionStringBuilder(connStr);
+        if (string.IsNullOrWhiteSpace(csb.Password))
+            throw new InvalidOperationException(
+                "ConnectionStrings:Postgres has no password. Set the full connection string via " +
+                "environment/user-secrets (see README) — a default password is deliberately not shipped.");
         await db.Database.MigrateAsync();
+    }
 }
 
 // Honor X-Forwarded-* from the nginx reverse proxy so OAuth redirect URIs and
@@ -128,15 +227,14 @@ var forwardedOptions = new ForwardedHeadersOptions
     ForwardedHeaders = ForwardedHeaders.XForwardedFor | ForwardedHeaders.XForwardedProto | ForwardedHeaders.XForwardedHost,
     ForwardLimit = app.Configuration.GetValue<int?>("ForwardedHeaders:ForwardLimit") ?? 1
 };
-forwardedOptions.KnownNetworks.Clear();
+forwardedOptions.KnownIPNetworks.Clear();
 forwardedOptions.KnownProxies.Clear();
 var trustedNetworks = app.Configuration.GetSection("ForwardedHeaders:KnownNetworks").Get<string[]>()
     ?? new[] { "10.0.0.0/8", "172.16.0.0/12", "192.168.0.0/16", "127.0.0.0/8", "::1/128" };
 foreach (var cidr in trustedNetworks)
 {
-    var parts = cidr.Split('/');
+    if (System.Net.IPNetwork.TryParse(cidr, out var network))
-    if (parts.Length == 2 && System.Net.IPAddress.TryParse(parts[0], out var prefix) && int.TryParse(parts[1], out var len))
+        forwardedOptions.KnownIPNetworks.Add(network);
-        forwardedOptions.KnownNetworks.Add(new Microsoft.AspNetCore.HttpOverrides.IPNetwork(prefix, len));
 }
 app.UseForwardedHeaders(forwardedOptions);
 
@@ -169,9 +267,22 @@ app.Use(async (ctx, next) =>
 app.UseSerilogRequestLogging();
 app.UseCors("frontend");
 app.UseAuthentication();
+// AUDIT H-2: after authentication so authenticated traffic partitions per-user; anonymous
+// traffic partitions per-IP. Endpoint policies ("auth", "expensive") apply via attributes.
+app.UseRateLimiter();
 app.UseAuthorization();
 app.MapControllers();
 
 app.Run();
 
 public partial class Program { }
+
+/// <summary>
+/// AUDIT M-1: absolute session lifetime check, extracted for unit testing. A session with no
+/// issued stamp (pre-dating this feature) is treated as expired — one forced re-login.
+/// </summary>
+public static class SessionLifetime
+{
+    public static bool IsExpired(string? issuedAtIso, DateTimeOffset now, TimeSpan maxAge)
+        => !DateTimeOffset.TryParse(issuedAtIso, out var issued) || now - issued > maxAge;
+}

src/InboxIntel.Api/appsettings.json

@@ -1,12 +1,18 @@
 {
   "ConnectionStrings": {
-    "Postgres": "Host=localhost;Port=5432;Database=inboxintel;Username=inboxintel;Password=inboxintel"
+    "Postgres": ""
   },
   "Database": {
     "AutoMigrate": true
   },
   "DataProtection": {
-    "KeyPath": "/keys"
+    "KeyPath": "/keys",
+    "CertificatePath": "",
+    "CertificatePassword": ""
+  },
+  "DataRetention": {
+    "PurgeTrashedAfterDays": 0,
+    "PurgeAllAfterDays": 0
   },
   "GoogleOAuth": {
     "ClientId": "",

src/InboxIntel.Application/Abstractions/IServices.cs

@@ -22,7 +22,6 @@ public interface IAnalyticsService
     Task<InboxHealthDto> GetInboxHealthAsync(Guid userId, CancellationToken ct = default);
     Task<IReadOnlyList<SenderStatDto>> GetTopSendersAsync(Guid userId, int take = 20, CancellationToken ct = default);
     Task<IReadOnlyList<TimeSeriesPointDto>> GetVolumeOverTimeAsync(Guid userId, int days = 90, CancellationToken ct = default);
-    Task<IReadOnlyList<HeatmapCellDto>> GetHeatmapAsync(Guid userId, CancellationToken ct = default);
     Task<IReadOnlyList<CategoryHeatmapCellDto>> GetCategoryHeatmapAsync(Guid userId, CancellationToken ct = default);
     Task<IReadOnlyList<AttachmentBreakdownDto>> GetAttachmentBreakdownAsync(Guid userId, CancellationToken ct = default);
     Task<SidebarCountsDto> GetSidebarCountsAsync(Guid userId, CancellationToken ct = default);
@@ -82,6 +81,25 @@ public interface IAiProvider
     Task<string> CompleteAsync(string systemPrompt, string userPrompt, CancellationToken ct = default);
 }
 
+/// <summary>
+/// Produces vector embeddings for text (the foundation for semantic search, near-duplicate
+/// detection, and "find similar"). Kept separate from <see cref="IAiProvider"/> because
+/// embeddings are a distinct capability with their own model. The Null implementation returns
+/// an empty vector and <see cref="IsAvailable"/> = false, so callers detect unavailability and
+/// fall back to lexical search — AI is never required for core functionality.
+/// </summary>
+public interface IEmbeddingProvider
+{
+    /// <summary>False for the Null provider (AI off / no embedding model configured).</summary>
+    bool IsAvailable { get; }
+
+    /// <summary>Embed a single text. Returns an empty array when unavailable.</summary>
+    Task<float[]> EmbedAsync(string text, CancellationToken ct = default);
+
+    /// <summary>Embed many texts, result aligned to input order. Empty list when unavailable.</summary>
+    Task<IReadOnlyList<float[]>> EmbedBatchAsync(IReadOnlyList<string> texts, CancellationToken ct = default);
+}
+
 public enum ExportFormat { Pdf, Csv, Json }
 
 public interface IExportService
@@ -101,3 +119,15 @@ public interface IDigestService
 {
     Task SendDigestAsync(Guid userId, CancellationToken ct = default);
 }
+
+/// <summary>System feature flags (fail-closed: unknown key = disabled).</summary>
+public interface IFeatureFlags
+{
+    Task<bool> IsEnabledAsync(string key, CancellationToken ct = default);
+}
+
+/// <summary>Policy gate for AI features: system flag AND the user's opt-in.</summary>
+public interface IAiGate
+{
+    Task<bool> IsAiEnabledForUserAsync(Guid userId, CancellationToken ct = default);
+}

src/InboxIntel.Application/DTOs/AnalyticsDtos.cs

@@ -12,7 +12,6 @@ public record InboxHealthDto(
 
 public record TimeSeriesPointDto(DateOnly Day, int Count);
 
-public record HeatmapCellDto(int DayOfWeek, int Hour, int Count);
 
 /// <summary>Email counts per category per day-of-week, for the category heatmap.</summary>
 public record CategoryHeatmapCellDto(string Category, int DayOfWeek, int Count);
@@ -39,6 +38,5 @@ public record DashboardSummaryDto(
     int UnreadEmails,
     IReadOnlyList<SenderStatDto> TopSenders,
     IReadOnlyList<TimeSeriesPointDto> VolumeOverTime,
-    IReadOnlyList<HeatmapCellDto> Heatmap,
     IReadOnlyList<AttachmentBreakdownDto> AttachmentBreakdown,
     long StorageEstimateBytes);

src/InboxIntel.Application/DTOs/EmailDtos.cs

@@ -2,6 +2,13 @@ using InboxIntel.Domain.Enums;
 
 namespace InboxIntel.Application.DTOs;
 
+/// <summary>
+/// Row summary for search/list results. <c>MatchHighlight</c> is a "why this matched" body
+/// fragment (ts_headline) with matched terms wrapped in U+E000/U+E001 sentinels — NOT HTML;
+/// the client renders them as escaped &lt;mark&gt; elements, so untrusted email content can
+/// never inject markup. Null unless the search had a free-text query; optional/last so other
+/// DTO constructors are unaffected.
+/// </summary>
 public record EmailSummaryDto(
     Guid Id,
     string GmailMessageId,
@@ -16,7 +23,8 @@ public record EmailSummaryDto(
     long SizeEstimateBytes,
     EmailCategory Category,
     bool HasListUnsubscribe,
-    bool SupportsOneClick);
+    bool SupportsOneClick,
+    string? MatchHighlight = null);
 
 /// <summary>Full single-email view, including body text, for the detail pane.</summary>
 public record EmailDetailDto(

src/InboxIntel.Application/DTOs/SearchDtos.cs

@@ -21,6 +21,19 @@ public record SearchRequestDto(
     bool? IsInInbox = null,
     bool? IsStarred = null,
     bool? IsTrashed = null,
+    bool? IsImportant = null,    // Gmail "important" marker — backs the Pinned smart folder
+    bool? IsReadLater = null,    // local Read Later marker
     string? GmailLabel = null,   // e.g. "SENT", "DRAFT", "SPAM"
+                                 // Emails carrying NONE of these Gmail labels (by GmailLabelId). Backs Archive, which
+                                 // must exclude Sent/Spam/Draft/Chat/Trash rather than just "not in inbox".
+    IReadOnlyList<string>? ExcludeGmailLabels = null,
+    // true = only emails with at least one USER label; false = only emails with NO user
+    // labels (the Unlabelled folder). System labels (INBOX/SENT/…) don't count.
+    bool? HasUserLabels = null,
     string? Category = null,     // EmailCategory name, e.g. "Finance"
-    long? MinSizeBytes = null);
+    long? MinSizeBytes = null,
+    // Keyset cursor for the date-ordered browse path (RECOMMENDATIONS #8): pass the last
+    // row's SentAtUtc+Id to fetch the next window without OFFSET (O(pageSize), not O(page)).
+    // When set, TotalCount is not recomputed (-1). Additive; offset paging still works.
+    DateTimeOffset? AfterSentAtUtc = null,
+    Guid? AfterId = null);

src/InboxIntel.Application/InboxIntel.Application.csproj

@@ -6,10 +6,13 @@
   <ItemGroup>
     <PackageReference Include="FluentValidation" Version="11.9.2" />
     <PackageReference Include="FluentValidation.DependencyInjectionExtensions" Version="11.9.2" />
-    <PackageReference Include="Microsoft.Extensions.DependencyInjection.Abstractions" Version="8.0.1" />
+    <PackageReference Include="Microsoft.Extensions.DependencyInjection.Abstractions" Version="10.0.9" />
     <!-- DbSet<> is exposed on IAppDbContext so the Application layer can query.
          Pinned to 8.0.4 to match the Npgsql provider's Relational dependency. -->
-    <PackageReference Include="Microsoft.EntityFrameworkCore" Version="8.0.4" />
+    <PackageReference Include="Microsoft.EntityFrameworkCore" Version="10.0.9" />
+    <!-- Transitive security pins: patch known .NET 8.0.0 advisories pulled in by EF Core. -->
+    <PackageReference Include="System.Text.Json" Version="10.0.9" />
+    <PackageReference Include="Microsoft.Extensions.Caching.Memory" Version="10.0.9" />
   </ItemGroup>
   <ItemGroup>
     <ProjectReference Include="..\InboxIntel.Domain\InboxIntel.Domain.csproj" />

src/InboxIntel.Domain/Entities/Email.cs

@@ -40,6 +40,10 @@ public class Email : AuditableEntity
     public bool IsTrashed { get; set; }
     public bool HasAttachments { get; set; }
 
+    /// <summary>Local "Read Later" marker (not a Gmail concept) — toggled by the user so the
+    /// Read Later smart folder shows only explicitly flagged mail, never everything.</summary>
+    public bool IsReadLater { get; set; }
+
     // Unsubscribe signals captured at parse time.
     public bool HasListUnsubscribe { get; set; }
     public string? ListUnsubscribeRaw { get; set; }
@@ -54,6 +58,13 @@ public class Email : AuditableEntity
     /// </summary>
     public NpgsqlTypes.NpgsqlTsVector? SearchVector { get; set; }
 
+    /// <summary>
+    /// Semantic-search embedding (pgvector, 768-dim for nomic-embed-text). Populated by the
+    /// embedding backfill worker when AI is enabled; null otherwise (search falls back to
+    /// lexical). Nullable so the InMemory test provider and AI-off deployments work unchanged.
+    /// </summary>
+    public Pgvector.Vector? Embedding { get; set; }
+
     public ICollection<EmailLabel> EmailLabels { get; set; } = new List<EmailLabel>();
     public ICollection<Attachment> Attachments { get; set; } = new List<Attachment>();
 }

src/InboxIntel.Domain/Entities/FeatureFlag.cs

@@ -0,0 +1,42 @@
+using InboxIntel.Domain.Common;
+
+namespace InboxIntel.Domain.Entities;
+
+/// <summary>
+/// System-wide feature flag (docs/discovery/multi-provider/04). The admin master switches:
+/// a disabled flag turns its feature off for EVERYONE regardless of user preferences.
+/// Reads are fail-closed — an unknown key counts as disabled.
+/// </summary>
+public class FeatureFlag : AuditableEntity
+{
+    /// <summary>Stable key, e.g. "ai.enabled", "provider.google".</summary>
+    public string Key { get; set; } = string.Empty;
+
+    public bool Enabled { get; set; }
+
+    /// <summary>True = a user preference may turn the feature OFF for themselves
+    /// (never on beyond the flag); false = system-only switch.</summary>
+    public bool UserOverridable { get; set; }
+
+    public string? Description { get; set; }
+}
+
+/// <summary>
+/// Per-user preferences (docs/discovery/multi-provider/04). One row per user, created
+/// lazily; absent row = defaults. AiOptIn defaults true so enabling the ai.enabled flag
+/// behaves exactly like today until a user opts out.
+/// </summary>
+public class UserSetting : AuditableEntity
+{
+    public Guid UserId { get; set; }
+    public User? User { get; set; }
+
+    /// <summary>"system" | "light" | "dark".</summary>
+    public string Theme { get; set; } = "dark";
+
+    /// <summary>Master per-user AI opt-in (effective only while ai.enabled is on).</summary>
+    public bool AiOptIn { get; set; } = true;
+
+    /// <summary>Free-form UI preferences (layout, density, notifications) as JSON.</summary>
+    public string? PreferencesJson { get; set; }
+}

src/InboxIntel.Domain/InboxIntel.Domain.csproj

@@ -4,7 +4,9 @@
     <AssemblyName>InboxIntel.Domain</AssemblyName>
   </PropertyGroup>
   <ItemGroup>
-    <!-- NpgsqlTypes.NpgsqlTsVector is used on the Email entity for FTS mapping. -->
+    <!-- NpgsqlTypes.NpgsqlTsVector (FTS) and Pgvector.Vector (semantic search) are used as
-    <PackageReference Include="Npgsql" Version="8.0.3" />
+         column types on the Email entity — same pragmatic precedent for both. -->
+    <PackageReference Include="Npgsql" Version="10.0.2" />
+    <PackageReference Include="Pgvector" Version="0.3.0" />
   </ItemGroup>
 </Project>

src/InboxIntel.Infrastructure/Ai/AiProviders.cs

@@ -2,6 +2,7 @@ using InboxIntel.Application.Abstractions;
 using InboxIntel.Domain.Enums;
 using InboxIntel.Infrastructure.Configuration;
 using Microsoft.Extensions.Options;
+using System.Linq;
 using System.Net.Http.Json;
 using System.Text.Json;
 
@@ -15,6 +16,53 @@ public class NullAiProvider : IAiProvider
         => Task.FromResult(string.Empty);
 }
 
+/// <summary>No-op embeddings used when AI is disabled. Returns empty vectors so callers fall
+/// back to lexical search.</summary>
+public class NullEmbeddingProvider : IEmbeddingProvider
+{
+    public bool IsAvailable => false;
+    public Task<float[]> EmbedAsync(string text, CancellationToken ct = default)
+        => Task.FromResult(Array.Empty<float>());
+    public Task<IReadOnlyList<float[]>> EmbedBatchAsync(IReadOnlyList<string> texts, CancellationToken ct = default)
+        => Task.FromResult<IReadOnlyList<float[]>>(Array.Empty<float[]>());
+}
+
+/// <summary>Local embeddings via Ollama's /api/embeddings endpoint (e.g. nomic-embed-text).</summary>
+public class OllamaEmbeddingProvider : IEmbeddingProvider
+{
+    private readonly HttpClient _http;
+    private readonly AiOptions _options;
+
+    public OllamaEmbeddingProvider(IHttpClientFactory factory, IOptions<AiOptions> options)
+    {
+        _options = options.Value;
+        _http = factory.CreateClient("ollama");
+        _http.BaseAddress = new Uri(_options.OllamaBaseUrl);
+    }
+
+    public bool IsAvailable => true;
+
+    public async Task<float[]> EmbedAsync(string text, CancellationToken ct = default)
+    {
+        var payload = new { model = _options.EmbeddingModel, prompt = text };
+        var resp = await _http.PostAsJsonAsync("/api/embeddings", payload, ct);
+        resp.EnsureSuccessStatusCode();
+        using var doc = JsonDocument.Parse(await resp.Content.ReadAsStringAsync(ct));
+        return doc.RootElement.GetProperty("embedding").EnumerateArray()
+            .Select(e => e.GetSingle()).ToArray();
+    }
+
+    // Ollama's /api/embeddings takes one prompt per call, so batch is a sequential loop.
+    // Kept behind the interface so a future batch endpoint is a drop-in swap.
+    public async Task<IReadOnlyList<float[]>> EmbedBatchAsync(IReadOnlyList<string> texts, CancellationToken ct = default)
+    {
+        var results = new List<float[]>(texts.Count);
+        foreach (var t in texts)
+            results.Add(await EmbedAsync(t, ct));
+        return results;
+    }
+}
+
 /// <summary>Local LLM via Ollama's /api/chat endpoint.</summary>
 public class OllamaProvider : IAiProvider
 {

src/InboxIntel.Infrastructure/Ai/EmbeddingBackfillWorker.cs

@@ -0,0 +1,104 @@
+using InboxIntel.Application.Abstractions;
+using InboxIntel.Infrastructure.Persistence;
+using Microsoft.EntityFrameworkCore;
+using Microsoft.Extensions.DependencyInjection;
+using Microsoft.Extensions.Hosting;
+using Microsoft.Extensions.Logging;
+
+namespace InboxIntel.Infrastructure.Ai;
+
+/// <summary>
+/// Fills <c>Email.Embedding</c> (pgvector) for semantic search, in small background batches
+/// so interactive requests are never starved (per docs/discovery/06: embeddings are the
+/// small always-on model; the batch pause keeps VRAM/CPU pressure low). Exits immediately
+/// when the embedding provider is unavailable (AI disabled / Ollama down) — semantic search
+/// simply stays dormant and lexical search is unaffected.
+/// </summary>
+public class EmbeddingBackfillWorker : BackgroundService
+{
+    private const int BatchSize = 32;
+    private static readonly TimeSpan BatchPause = TimeSpan.FromSeconds(2);
+    private static readonly TimeSpan IdleRescan = TimeSpan.FromMinutes(15);
+
+    private readonly IServiceScopeFactory _scopeFactory;
+    private readonly ILogger<EmbeddingBackfillWorker> _logger;
+
+    public EmbeddingBackfillWorker(IServiceScopeFactory scopeFactory, ILogger<EmbeddingBackfillWorker> logger)
+    {
+        _scopeFactory = scopeFactory;
+        _logger = logger;
+    }
+
+    protected override async Task ExecuteAsync(CancellationToken stoppingToken)
+    {
+        // Provider availability is fixed by configuration for the process lifetime.
+        using (var probe = _scopeFactory.CreateScope())
+        {
+            if (!probe.ServiceProvider.GetRequiredService<IEmbeddingProvider>().IsAvailable)
+            {
+                _logger.LogDebug("EmbeddingBackfillWorker idle: no embedding provider (AI disabled).");
+                return;
+            }
+        }
+
+        _logger.LogInformation("EmbeddingBackfillWorker started (batch {Batch}, pause {Pause}s)",
+            BatchSize, BatchPause.TotalSeconds);
+
+        while (!stoppingToken.IsCancellationRequested)
+        {
+            int processed;
+            try
+            {
+                processed = await ProcessBatchAsync(stoppingToken);
+            }
+            catch (OperationCanceledException) when (stoppingToken.IsCancellationRequested) { break; }
+            catch (Exception ex)
+            {
+                // Ollama hiccups must never crash the host; back off and retry.
+                _logger.LogWarning(ex, "Embedding batch failed; retrying after idle pause.");
+                processed = 0;
+            }
+
+            await Task.Delay(processed > 0 ? BatchPause : IdleRescan, stoppingToken);
+        }
+    }
+
+    /// <summary>Embeds one batch. Public-ish (internal) for direct testing.</summary>
+    internal async Task<int> ProcessBatchAsync(CancellationToken ct)
+    {
+        using var scope = _scopeFactory.CreateScope();
+        var db = scope.ServiceProvider.GetRequiredService<AppDbContext>();
+        var embeddings = scope.ServiceProvider.GetRequiredService<IEmbeddingProvider>();
+
+        var batch = await db.Emails
+            .Where(e => e.Embedding == null)
+            .OrderByDescending(e => e.SentAtUtc) // newest mail becomes searchable first
+            .Take(BatchSize)
+            .ToListAsync(ct);
+        if (batch.Count == 0) return 0;
+
+        // Subject + snippet is the semantic core; bodies are noisy (signatures, quoting)
+        // and slow to embed. Truncate defensively to keep well inside the model context.
+        var texts = batch
+            .Select(e => Truncate($"{e.Subject}\n{e.Snippet ?? e.BodyText}", 2000))
+            .ToList();
+        var vectors = await embeddings.EmbedBatchAsync(texts, ct);
+        if (vectors.Count != batch.Count)
+        {
+            _logger.LogWarning("Embedding batch returned {Got} vectors for {Want} emails; skipping batch.",
+                vectors.Count, batch.Count);
+            return 0;
+        }
+
+        for (var i = 0; i < batch.Count; i++)
+        {
+            if (vectors[i].Length == 0) continue; // provider soft-failure for one item
+            batch[i].Embedding = new Pgvector.Vector(vectors[i]);
+        }
+        await db.SaveChangesAsync(ct);
+        _logger.LogDebug("Embedded {Count} emails", batch.Count);
+        return batch.Count;
+    }
+
+    private static string Truncate(string s, int max) => s.Length <= max ? s : s[..max];
+}

src/InboxIntel.Infrastructure/Analytics/AnalyticsService.cs

@@ -16,11 +16,10 @@ public class AnalyticsService : IAnalyticsService
         var health = await GetInboxHealthAsync(userId, ct);
         var top = await GetTopSendersAsync(userId, 10, ct);
         var volume = await GetVolumeOverTimeAsync(userId, 90, ct);
-        var heatmap = await GetHeatmapAsync(userId, ct);
         var attachments = await GetAttachmentBreakdownAsync(userId, ct);
 
         return new DashboardSummaryDto(
-            health, health.TotalEmails, health.UnreadEmails, top, volume, heatmap, attachments,
+            health, health.TotalEmails, health.UnreadEmails, top, volume, attachments,
             health.EstimatedStorageBytes);
     }
 
@@ -75,19 +74,6 @@ public class AnalyticsService : IAnalyticsService
             .ToList();
     }
 
-    public async Task<IReadOnlyList<HeatmapCellDto>> GetHeatmapAsync(Guid userId, CancellationToken ct = default)
-    {
-        var raw = await _db.Emails
-            .Where(e => e.UserId == userId)
-            .Select(e => new { e.SentAtUtc })
-            .ToListAsync(ct);
-
-        return raw
-            .GroupBy(x => new { Dow = (int)x.SentAtUtc.DayOfWeek, Hour = x.SentAtUtc.Hour })
-            .Select(g => new HeatmapCellDto(g.Key.Dow, g.Key.Hour, g.Count()))
-            .ToList();
-    }
-
     public async Task<IReadOnlyList<CategoryHeatmapCellDto>> GetCategoryHeatmapAsync(Guid userId, CancellationToken ct = default)
     {
         var raw = await _db.Emails
@@ -133,13 +119,13 @@ public class AnalyticsService : IAnalyticsService
         var emails = _db.Emails.Where(e => e.UserId == userId);
 
         var allMail = await emails.CountAsync(ct);
-        var inbox   = await emails.CountAsync(e => e.IsInInbox, ct);
+        var inbox = await emails.CountAsync(e => e.IsInInbox, ct);
-        var unread  = await emails.CountAsync(e => e.IsUnread, ct);
+        var unread = await emails.CountAsync(e => e.IsUnread, ct);
         var starred = await emails.CountAsync(e => e.IsStarred, ct);
-        var trash   = await emails.CountAsync(e => e.IsTrashed, ct);
+        var trash = await emails.CountAsync(e => e.IsTrashed, ct);
-        var large   = await emails.CountAsync(e => e.SizeEstimateBytes > 5_000_000, ct);
+        var large = await emails.CountAsync(e => e.SizeEstimateBytes > 5_000_000, ct);
-        var cutoff  = DateTimeOffset.UtcNow.AddYears(-1);
+        var cutoff = DateTimeOffset.UtcNow.AddYears(-1);
-        var old     = await emails.CountAsync(e => e.SentAtUtc < cutoff, ct);
+        var old = await emails.CountAsync(e => e.SentAtUtc < cutoff, ct);
 
         // Label-backed counts (SENT / DRAFT / SPAM are system Gmail labels)
         async Task<int> LabelCount(string gmailId)
@@ -153,9 +139,9 @@ public class AnalyticsService : IAnalyticsService
                 : 0;
         }
 
-        var sent   = await LabelCount("SENT");
+        var sent = await LabelCount("SENT");
         var drafts = await LabelCount("DRAFT");
-        var spam   = await LabelCount("SPAM");
+        var spam = await LabelCount("SPAM");
 
         // Category → smart-folder slug mapping
         var catCounts = await emails
@@ -167,31 +153,31 @@ public class AnalyticsService : IAnalyticsService
 
         var smartFolders = new Dictionary<string, int>
         {
-            ["automated"]    = Cat(EmailCategory.Notification),
+            ["automated"] = Cat(EmailCategory.Notification),
-            ["finance"]      = Cat(EmailCategory.Finance),
+            ["finance"] = Cat(EmailCategory.Finance),
-            ["social"]       = Cat(EmailCategory.Social),
+            ["social"] = Cat(EmailCategory.Social),
-            ["shopping"]     = Cat(EmailCategory.Shopping) + Cat(EmailCategory.Promotional),
+            ["shopping"] = Cat(EmailCategory.Shopping) + Cat(EmailCategory.Promotional),
-            ["noreply"]      = Cat(EmailCategory.Notification),
+            ["noreply"] = Cat(EmailCategory.Notification),
-            ["gaming"]       = Cat(EmailCategory.Gaming),
+            ["gaming"] = Cat(EmailCategory.Gaming),
-            ["sales"]        = Cat(EmailCategory.SeasonalSales),
+            ["sales"] = Cat(EmailCategory.SeasonalSales),
-            ["ridesharing"]  = Cat(EmailCategory.RideSharing),
+            ["ridesharing"] = Cat(EmailCategory.RideSharing),
-            ["food"]         = Cat(EmailCategory.FoodDelivery),
+            ["food"] = Cat(EmailCategory.FoodDelivery),
-            ["wellness"]     = Cat(EmailCategory.Wellness),
+            ["wellness"] = Cat(EmailCategory.Wellness),
             // New categories
-            ["travel"]        = Cat(EmailCategory.Travel),
+            ["travel"] = Cat(EmailCategory.Travel),
             ["subscriptions"] = Cat(EmailCategory.Subscriptions),
-            ["parcels"]       = Cat(EmailCategory.Parcels),
+            ["parcels"] = Cat(EmailCategory.Parcels),
-            ["recruitment"]   = Cat(EmailCategory.Recruitment),
+            ["recruitment"] = Cat(EmailCategory.Recruitment),
-            ["events"]        = Cat(EmailCategory.Events),
+            ["events"] = Cat(EmailCategory.Events),
-            ["security"]      = Cat(EmailCategory.SecurityAlerts),
+            ["security"] = Cat(EmailCategory.SecurityAlerts),
-            ["healthcare"]    = Cat(EmailCategory.Healthcare),
+            ["healthcare"] = Cat(EmailCategory.Healthcare),
-            ["education"]     = Cat(EmailCategory.Education),
+            ["education"] = Cat(EmailCategory.Education),
-            ["news"]          = Cat(EmailCategory.NewsMedia),
+            ["news"] = Cat(EmailCategory.NewsMedia),
-            ["property"]      = Cat(EmailCategory.PropertyUtilities),
+            ["property"] = Cat(EmailCategory.PropertyUtilities),
-            ["charity"]       = Cat(EmailCategory.Charity),
+            ["charity"] = Cat(EmailCategory.Charity),
-            ["government"]    = Cat(EmailCategory.Government),
+            ["government"] = Cat(EmailCategory.Government),
-            ["crypto"]        = Cat(EmailCategory.CryptoInvesting),
+            ["crypto"] = Cat(EmailCategory.CryptoInvesting),
-            ["family"]        = Cat(EmailCategory.FamilySchool),
+            ["family"] = Cat(EmailCategory.FamilySchool),
         };
 
         return new SidebarCountsDto(inbox, allMail, unread, starred, sent, drafts, trash, spam, large, old, smartFolders);

src/InboxIntel.Infrastructure/Configuration/Options.cs

@@ -43,6 +43,8 @@ public class AiOptions
     // Ollama (local)
     public string OllamaBaseUrl { get; set; } = "http://localhost:11434";
     public string OllamaModel { get; set; } = "llama3.1";
+    // Embedding model for semantic search (per docs/discovery/06). Small, always-on when local.
+    public string EmbeddingModel { get; set; } = "nomic-embed-text";
 
     // OpenAI (cloud, optional)
     public string OpenAiApiKey { get; set; } = string.Empty;
@@ -70,3 +72,13 @@ public class DigestOptions
     /// <summary>Hour (UTC) the background worker checks for due digests.</summary>
     public int SendHourUtc { get; set; } = 8;
 }
+
+/// <summary>AUDIT H-3: opt-in local data retention. 0 = disabled (keep forever).</summary>
+public class DataRetentionOptions
+{
+    public const string SectionName = "DataRetention";
+    /// <summary>Purge locally stored emails flagged Trashed older than this many days.</summary>
+    public int PurgeTrashedAfterDays { get; set; } = 0;
+    /// <summary>Purge ALL locally stored emails older than this many days (local copy only).</summary>
+    public int PurgeAllAfterDays { get; set; } = 0;
+}

src/InboxIntel.Infrastructure/DependencyInjection.cs

@@ -21,10 +21,11 @@ public static class DependencyInjection
 {
     public static IServiceCollection AddInfrastructure(this IServiceCollection services, IConfiguration config)
     {
-        // EF Core / PostgreSQL
+        // EF Core / PostgreSQL. UseVector() enables pgvector mapping for the semantic-search
+        // embedding column (requires the 'vector' extension — added by the AddEmbeddingColumn migration).
         services.AddDbContext<AppDbContext>(opt =>
             opt.UseNpgsql(config.GetConnectionString("Postgres"),
-                npg => npg.MigrationsAssembly(typeof(AppDbContext).Assembly.FullName)));
+                npg => npg.MigrationsAssembly(typeof(AppDbContext).Assembly.FullName).UseVector()));
         services.AddScoped<IAppDbContext>(sp => sp.GetRequiredService<AppDbContext>());
 
         // Options
@@ -59,6 +60,11 @@ public static class DependencyInjection
         services.AddScoped<IDigestService, DigestService>();
         services.AddHostedService<DigestWorker>();
 
+        // AUDIT H-3: opt-in local data retention (worker no-ops while disabled).
+        services.Configure<DataRetentionOptions>(config.GetSection(DataRetentionOptions.SectionName));
+        services.AddScoped<Retention.RetentionService>();
+        services.AddHostedService<Retention.RetentionWorker>();
+
         // HTTP clients
         // V-01: do NOT follow redirects — a validated external URL must not be able to
         // 3xx-redirect into an internal target after SafeHttpGuard has checked it.
@@ -67,15 +73,33 @@ public static class DependencyInjection
         services.AddHttpClient("ollama");
         services.AddHttpClient("openai");
 
-        // AI provider selected by configured mode.
+        // AI provider selected by configured mode. Embeddings come from Ollama when local,
+        // otherwise the Null provider (empty vectors) so semantic features degrade to lexical.
         var aiMode = config.GetSection(AiOptions.SectionName).GetValue<AiProviderMode>("Mode");
         switch (aiMode)
         {
-            case AiProviderMode.LocalOllama: services.AddScoped<IAiProvider, OllamaProvider>(); break;
+            case AiProviderMode.LocalOllama:
-            case AiProviderMode.CloudOpenAi: services.AddScoped<IAiProvider, OpenAiProvider>(); break;
+                services.AddScoped<IAiProvider, OllamaProvider>();
-            default: services.AddScoped<IAiProvider, NullAiProvider>(); break;
+                services.AddScoped<IEmbeddingProvider, OllamaEmbeddingProvider>();
+                break;
+            case AiProviderMode.CloudOpenAi:
+                services.AddScoped<IAiProvider, OpenAiProvider>();
+                services.AddScoped<IEmbeddingProvider, NullEmbeddingProvider>(); // OpenAI embeddings: future
+                break;
+            default:
+                services.AddScoped<IAiProvider, NullAiProvider>();
+                services.AddScoped<IEmbeddingProvider, NullEmbeddingProvider>();
+                break;
         }
         services.AddScoped<IAiService, AiService>();
+        // Feature flags + AI policy gate (docs/discovery/multi-provider/04). Cached 15s,
+        // fail-closed. Admin toggle surface arrives with the multi-provider admin phase.
+        services.AddMemoryCache();
+        services.AddScoped<IFeatureFlags, Features.FeatureFlagService>();
+        services.AddScoped<IAiGate, Features.AiGate>();
+        // Semantic search: fills Email.Embedding in the background; no-ops when the
+        // embedding provider is unavailable (AI disabled), so lexical search is unaffected.
+        services.AddHostedService<EmbeddingBackfillWorker>();
 
         // Background worker (daily incremental sync + aggregate refresh)
         services.AddHostedService<GmailSyncWorker>();

src/InboxIntel.Infrastructure/Features/FeatureFlagService.cs

@@ -0,0 +1,72 @@
+using InboxIntel.Application.Abstractions;
+using InboxIntel.Infrastructure.Persistence;
+using Microsoft.EntityFrameworkCore;
+using Microsoft.Extensions.Caching.Memory;
+
+namespace InboxIntel.Infrastructure.Features;
+
+/// <summary>
+/// Flag evaluation (docs/discovery/multi-provider/04). DB-backed with a short cache so an
+/// admin toggle takes effect within seconds and per-request reads stay free.
+/// FAIL-CLOSED: unknown keys and read errors evaluate to disabled.
+/// </summary>
+public class FeatureFlagService : IFeatureFlags
+{
+    private static readonly TimeSpan CacheTtl = TimeSpan.FromSeconds(15);
+    private readonly AppDbContext _db;
+    private readonly IMemoryCache _cache;
+
+    public FeatureFlagService(AppDbContext db, IMemoryCache cache)
+    {
+        _db = db;
+        _cache = cache;
+    }
+
+    public async Task<bool> IsEnabledAsync(string key, CancellationToken ct = default)
+    {
+        try
+        {
+            var flags = await _cache.GetOrCreateAsync("feature-flags", async e =>
+            {
+                e.AbsoluteExpirationRelativeToNow = CacheTtl;
+                return await _db.FeatureFlags.AsNoTracking()
+                    .ToDictionaryAsync(f => f.Key, f => f.Enabled, ct);
+            });
+            return flags is not null && flags.TryGetValue(key, out var enabled) && enabled;
+        }
+        catch
+        {
+            return false; // fail closed
+        }
+    }
+}
+
+/// <summary>
+/// The AI gate (the audit/design requirement that AI is governed by a FLAG, not only user
+/// settings): effective AI = ai.enabled (admin, global) AND the user's opt-in (default true,
+/// only consulted while the flag is on). Callers still check provider availability
+/// (IAiService.IsEnabled / IEmbeddingProvider.IsAvailable) — this gate is policy, not plumbing.
+/// </summary>
+public class AiGate : IAiGate
+{
+    public const string MasterFlag = "ai.enabled";
+    private readonly IFeatureFlags _flags;
+    private readonly AppDbContext _db;
+
+    public AiGate(IFeatureFlags flags, AppDbContext db)
+    {
+        _flags = flags;
+        _db = db;
+    }
+
+    public async Task<bool> IsAiEnabledForUserAsync(Guid userId, CancellationToken ct = default)
+    {
+        if (!await _flags.IsEnabledAsync(MasterFlag, ct)) return false;
+        // Absent settings row = default opt-in true.
+        var optIn = await _db.UserSettings.AsNoTracking()
+            .Where(s => s.UserId == userId)
+            .Select(s => (bool?)s.AiOptIn)
+            .FirstOrDefaultAsync(ct);
+        return optIn ?? true;
+    }
+}

src/InboxIntel.Infrastructure/Gmail/GmailService.cs

@@ -95,7 +95,9 @@ public class GmailApiService : IGmailService
             var req = client.Users.Messages.List("me");
             req.MaxResults = _options.PageSize;
             req.PageToken = pageToken;
-            req.IncludeSpamTrash = false;
+            // Include spam & trash so those folders aren't structurally empty; their state is
+            // captured via the SPAM/TRASH labels (IsTrashed + EmailLabels) during upsert.
+            req.IncludeSpamTrash = true;
             return await req.ExecuteAsync(token);
         }, ct);