feat: in-process HTTP search; /get /stats /audit /ingest.

bin/brain/serve.go (ladybug tags) calls internal/brain instead of exec. HTTP tests inject a fake API so CI stays cgo-free. ExecSearcher remains the fallback when the binary is built without system_ladybug.
feat: brain/index.go shebang; mail import is not a brain write (D14). (#12 )
2026-08-13 17:50:46 +01:00 · 2026-08-13 17:46:25 +01:00 · 2026-08-13 17:31:03 +01:00 · 2026-08-13 17:23:40 +01:00 · 2026-08-13 14:32:21 +01:00 · 2026-08-13 14:26:54 +01:00
127 changed files with 11640 additions and 410 deletions
@@ -10,5 +10,4 @@ __pycache__
 .cache
 .secrets
 .skills-tmp
-serve/serve
 docs/.build
@@ -19,6 +19,10 @@ jobs:
        with:
          fetch-depth: 0

+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+
      - name: Install uv
        uses: astral-sh/setup-uv@v6
        with:
@@ -31,18 +35,20 @@ jobs:
        run: |
          bash -n bin/db/psql-yq
          bash -n bin/db/ssh-tunnel
-          bash -n bin/kb-watch
          bash -n bin/docker-entrypoint
+          bash -n bin/kb/search

      - name: Python unit tests (offline, vendored tools)
        run: |
-          uv run python -m unittest discover -s tools -t .
+          uv run python -m unittest discover -s bin/tools -t .

-      - name: Go serve tests (async, goroutine-bounded)
+      - name: Go tests (root module, no ladybug cgo)
        run: |
          go vet ./...
          go test ./... -count=1
-        working-directory: serve
+
+      - name: brain ranking tests (no cgo / no ladybug)
+        run: go test ./internal/brain/rank -count=1

      - name: facts/audit self (lexicon consistency, no network)
        run: |
@@ -8,4 +8,6 @@ __pycache__/
 .DS_Store
 *.env
 .env
-.secrets/
+.secrets/
+lib-ladybug/
+go.work.local
@@ -24,7 +24,7 @@ Read first: [PLAN](PLAN.md) → [docs](docs/).
 2. **Read-only data sources.** Ladybug `var/kb.lbug` and Postgres are opened
   read-only for queries. Index rebuilds write to `var/` (gitignored).
 3. **PII.** `brain-test`, `cs_brain` client data is never read or quoted.
-4. **No main pushes.** Feature branches + PR via `gh`; CI must be green.
+4. **No main pushes.** Feature branches + GitHub PR (`gh`); CI (Actions) must be green. Work board: [Gitea issues](https://git.produktor.io/eSlider/2dph/issues).
 5. **TDD.** Failing test before tool code. Unit tests run offline against
   fixtures; network/db calls are wrapped.
 6. **docs reflect behaviour.** Any change updates `docs/` + `PLAN.md` status.
@@ -35,22 +35,53 @@ Read first: [PLAN](PLAN.md) → [docs](docs/).
 PLAN.md       decisions + execution + open questions
 docs/         published docs
 skills/       in-project agent skills (vendored, no external links)
-bin/          self-describing tools bin/{subject}/{method} (shebang)
-bin/kb-watch  corpus watcher (mtimes, no inotify deps)
+bin/          self-describing tools bin/{subject}/{method}.go (shebang)
+bin/brain/    search.go serve.go index.go get.go stats.go eval.go watch.go
+bin/chats/    sync.go import.go facts.go apply.go; libs in internal/chats
+bin/mail/     sync.go import.go (index_mail → brain/index.go)
+bin/markdown/ import.go (mistune leafs)
+bin/postgres/ query.go (read-only YAML)
+internal/     shared Go (brain/rank is cgo-free; chats parsers too)
+bin/watch/    corpus watcher (used by bin/brain/watch.go)
+bin/tools/    vendored python libs behind bin/* (kblib, yamlout, websearch)
 bin/docker-entrypoint  container entrypoint (brain index|search|serve|watch)
-serve/        async Go HTTP server (goroutines, bounded worker pool)
-tools/        vendored python libs behind bin/* (yamlout, websearch)
 compose.yaml  docker composition (root level, not docker/)
-Dockerfile    multi-stage: python deps + static Go serve
-var/          kb.lbug, caches (gitignored)
+Dockerfile    multi-stage: python deps + static Go binaries
+var/          kb.lbug, var/mail/*, caches (gitignored)
 .venv/        ladybug + model2vec + mistune
 ```

+## Mail pipeline
+
+```bash
+bin/mail/sync.go --source onlyoffice,gmail --workers 8 --out var/mail   # raw message.json + attachments
+bin/mail/sync.go --source gmail --query 'from:example.com' --out var/mail  # Gmail search (default in:inbox)
+bin/mail/import.go --from-raw var/mail                                  # message.json → message.md (convert only)
+bin/brain/index.go --rebuild                                            # rebuild brain incl. all mail (fresh DB)
+```
+
+- `sync` (Go) downloads messages + attachments; Gmail uses paginated list +
+  `body.attachmentId` (not partId) for attachments.
+- `import` converts body + attachments to markdown. PDFs use poppler
+  `pdftotext -layout` fast path (~15ms); textless/scanned PDFs fall back to
+  docling (isolated subprocess — its native onnx can segfault the parent).
+  Conversion never touches the brain DB (crash safety).
+- `index_mail` is a deprecation shim for `bin/brain/index.go --rebuild`. Ladybug
+  corrupts its WAL when brand-new leafs are bulk-inserted while FTS/vector
+  indexes exist; a fresh DB with indexes created last is the only safe path.
+  Keep conversion + indexing separate so a conversion crash can't leave the
+  DB mid-transaction.
+
 ## Tools

 ```bash
 bin/facts/audit ["self"|"facts"|"info"|"stale"]   # 2-source + staleness gate
-bin/kb/search "query" [--hop N] [--repo X]        # deduction search → YAML
+bin/facts/crm [--dry-run]                         # proof person↔company/company↔project (ooCRM × corpus SoT)
+bin/kb/search "query" [--repo X]                  # deprecated wrapper → bin/brain/search.go
+bin/brain/search.go "query" [--root facts|info]   # deduction search → YAML
+bin/brain/get.go <id> [--body]
+bin/markdown/import.go [dir]                      # mistune leaves → YAML
+bin/postgres/query.go --profile onlyoffice -c 'SELECT 1'
 bin/md/tables                                     # what the graph holds → YAML
 bin/brain/deduce "question"                       # thinking wrapper
 ```
@@ -58,6 +89,19 @@ bin/brain/deduce "question"                       # thinking wrapper
 Never start a shell command with `cd` — use the tool working-directory
 parameter. Search before reading whole files.

+## GitHub safety rules (ABSOLUTE — never violate)
+
+1. **No absolute paths in committed files.** Replace `/mnt/`, `/home/<user>/`,
+   `/Users/<user>/` with env vars (`$HOME`, `$PROJECTS_ROOT`, `$DOCS_BASE`).
+2. **No PII in commits.** No real names, phones, emails of third parties.
+   Test data must be synthetic (Alice, Bob, Charlie, Diana, example.com).
+3. **No credentials/secrets in commits.** API keys, tokens, passwords, session
+   strings, phone numbers only in gitignored `.env` files, referenced by path.
+4. **Curasoft, edelweiss — no files, no mentions.** Remove all traces if found.
+5. **Check git history before push.** If any commit contains leaks, rewrite
+   history (rebase + force push) AND delete affected GitHub releases/tags.
+6. **`docs/chat-import-plan.md`** — reference Gitea issue, never embed secrets.
+
 ## Communication

 Same tone as the corpus: plain, lists, no hype. Sign-off `Andriy Oblivantsev`.
@@ -14,23 +14,27 @@ COPY requirements.lock.txt /tmp/requirements.lock.txt
 RUN python -m pip install --no-cache-dir -r /tmp/requirements.lock.txt \
    && rm /tmp/requirements.lock.txt

-# Go serve: static binary, no interpreter at runtime
-FROM golang:1.25 AS serve-build
-WORKDIR /src/serve
-COPY serve/go.mod serve/go.sum* ./
-COPY serve .
-RUN CGO_ENABLED=0 go build -o /serve -ldflags="-s -w" .
+# Go services: static binaries, no interpreter at runtime
+FROM golang:1.25 AS go-build
+WORKDIR /src
+COPY go.mod ./
+COPY bin/server ./bin/server
+COPY bin/watch ./bin/watch
+RUN CGO_ENABLED=0 go build -o /serve ./bin/server \
+    && CGO_ENABLED=0 go build -o /watch ./bin/watch

-# runtime: python toolchain + Go server
+# runtime: python toolchain + Go services
 FROM base
 COPY . .
-COPY --from=serve-build /serve /app/serve/serve
-RUN chmod +x /app/bin/kb-watch /app/bin/docker-entrypoint \
+COPY --from=go-build /serve /app/bin/serve
+COPY --from=go-build /watch /app/bin/watch
+RUN chmod +x /app/bin/docker-entrypoint \
    && chown -R 2dph:2dph /app
 USER 2dph

 ENV PATH="/app/bin:${PATH}" \
-    KB_PY=python3
+    KB_PY=python3 \
+    KB_ROOT=/app
 HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
    CMD python -c "import model2vec, ladybug, mistune; print('ok')" || exit 1

@@ -25,11 +25,11 @@ detective method: **a fact needs ≥2 independent sources or it is
 | # | Question | Answer |
 |---|----------|--------|
 | D1 | RAG corpus | ops stack (chat, onlyoffice, gitea/NPM, searchxng, observability, ai-bot, mcp-servers, `~/.ssh/config`) + portfolio. Exclude `office.dev` + jobs/applications. |
-| D2 | skill merging | integrate skills **in this project** `skills/`; skip gitea / brain-detective-depe ndent skills. |
-| D3 | web search | import `web-search`, retire local `searxng-ops`. Vendored here, no remote link. |
+| D2 | skill merging | integrate skills **in this project** `skills/`; skip gitea / brain-dependent skills. |
+| D3 | web search | Vendored client; SearXNG URL is config. Optional Compose instance (sanitized settings). Do not run a second copy on a host that already has one. Empty/`throttled` ≠ “nothing exists”. |
 | D4 | embeddings | **model2vec** `minishlab/potion-multilingual-128M` instead of embeddinggemma. |
 | D5 | parser | **mistune** for MD → leaf extraction (duckdb-md documented as future optional SQL/export layer, not v1). |
-| D6 | graph engine | **LadybugDB** (Kuzu successor, MIT, embedded, native FTS+vector+Cypher). Python binding for `bin/*`; Go shebang for golang tools. |
+| D6 | graph engine | **LadybugDB**. Go is the service (`bin/brain/search.go`, `bin/brain/serve.go` in-process, `internal/brain`); Python remains for index/write until the Go write path is safe. |
 | D7 | db access | `db-yaml`/`psql-yq`-style, read-only, YAML out. OnlyOffice Postgres via SSH tunnel (`127.0.0.1:5433`). |
 | D8 | evidence | detective method: ≥2 independent sources or `(not confirmed)`. Auto-pair docker ps × compose × ssh-config × docs. |
 | D9 | facts/goal model | Who / What / How / Where / When + evidence + confidence on every edge. |
@@ -37,9 +37,11 @@ detective method: **a fact needs ≥2 independent sources or it is
 | D11 | strong/weak | `root` column: `facts` (strong) vs `info` (weak). Answer is `confirmed` only from facts root. |
 | D12 | transactional | facts and info split by root but **written in the same Ladybug transaction (ACID)** on every write. |
 | D13 | portfolio | start graph `(Person:eslider)-[:HAS]->(Portfolio)`, associate other natural/juristic persons later. |
-| D14 | tooling style | `bin/{subject}/{method}` self-describing: shebang line 1, usage comment from line 2. Go shebang: `///usr/bin/env go run "$0" "$@"; exit`. |
-| D15 | repo | GitHub `eSlider/2dph`, public (like sibling repos), push/commit via `gh`, TDD + commit every change, CI/CD. |
+| D14 | tooling style | `bin/{subject}/{method}.go` shebang (e.g. `bin/brain/search.go`). Shared code in `internal/`. One root `go.mod` + `go.work`. No `bin/*/main.go`, no nested modules. |
+| D15 | repo | Gitea [`eSlider/2dph`](https://git.produktor.io/eSlider/2dph) is origin + [issues](https://git.produktor.io/eSlider/2dph/issues). GitHub `eSlider/2dph` is the public clone (PRs + Actions CI). No direct `main` pushes. TDD → PR → CI green → merge. |
 | D16 | contradictions | ≥2 yes vs ≥2 no → unrelated sources conflict → hypothesis → `(not confirmed)`. Resolution (authority, staleness adjudication) = **v2**, tracked as open question. |
+| D17 | assertion gate | Fact-check every *claim* (facts → info → live sources → web), not every edit. Missing graph ≠ “does not exist”. |
+| D18 | reasoner | Pluggable OpenAI-compatible URL. RAM: Qwen3.5-9B. Quality: Bonsai-27B or Qwen3.6-27B. No official Qwen3.6-9B. |

 ## Architecture

@@ -51,10 +53,17 @@ detective method: **a fact needs ≥2 independent sources or it is
  bin/
    facts/extract           auto-pair 2 sources → lexicon yaml + graph
    facts/audit             ["self"|"facts"|"info"|"stale"] 2-source + staleness gate
-    kb/index                build FTS + HNSW from corpus
-    kb/search               deduction: facts → info → web-search; --hop N
-    kb/get  kb/stats  kb/eval
-    md/import  md/select  md/tables  md/gaps     (mistune)
+    kb/index                Python write path (called by bin/brain/index.go)
+    brain/index.go          rebuild FTS + HNSW (incl. --with-mail)
+    brain/get.go stats.go eval.go watch.go
+    brain/search.go         deduction: facts → info → web-search
+    brain/serve.go          HTTP API in-process (internal/httpapi + internal/brain)
+    mail/import.go          JSON → markdown (no brain write)
+    markdown/import.go      mistune leaves
+    postgres/query.go       read-only YAML (wraps bin/db/psql-yq)
+    chats/sync.go import.go facts.go apply.go
+                            (libs in internal/chats; no chats index)
+    md/import               (deprecated; bin/markdown/import.go)
    brain/extract  brain/audit   brain/deduce    (thinking wrapper)
    web/search               (vendored)
    db/psql-yq               (vendored)
@@ -93,18 +102,36 @@ Common props on every node/edge: `root`, `confidence`, `evidence[]`, `how`,

 - OQ1: mutually-contradicting evidence — how to resolve (authority weighting,
  temporal freshness, audit adjudication).
- OQ2: OCR pipeline for pdfs/images/docs (late phase).
+- OQ2: OCR pipeline for pdfs/images/docs — mostly solved: poppler pdftotext
+  fast-path for born-digital PDFs, docling fallback for the ~5% textless ones.
 - OQ3: optional duckdb-md layer for `SELECT … FORMAT MARKDOWN` export/write-back.
+- OQ4: YAML-first storage for leafs — deferred: JSON is ~10x faster to
+  serialize and unambiguous; YAML only where humans edit files.
+
+## Mail pipeline (done)
+
+1. `bin/mail/sync.go` (Go, 8 workers) — paginated Gmail/OnlyOffice download.
+   Gmail attachments key off `body.attachmentId`, not MIME `partId`.
+2. `bin/mail/import.go --from-raw` — message.json → message.md; PDFs via
+   `pdftotext -layout` (~15ms) with docling subprocess fallback; ICS sidecars
+   Latin-1→UTF-8 normalized.
+3. `bin/brain/index.go --rebuild` — fresh rebuild (repo corpus + mail) because ladybug
+   corrupts its WAL on bulk-insert into an already-indexed DB. Conversion and
+   indexing stay separate for crash safety. `bin/mail/index_mail` is a
+   deprecation shim.
+4. Result: 17,835 messages → 28,918 info leafs, FTS + HNSW healthy, searchable
+   via `bin/brain/search.go`.

 ## CI/CD pipeline (D15)

 `.github/workflows/ci.yml`:

-1. go vet + go test ./... (Go tools)
-2. python -m unittest discover + pytest (Py tools)
-3. bin/facts/audit self  (lexicon internal consistency)
-4. bin/kb/eval            (recall@5 ≥ 0.95, gates index regressions)
-5. md-docs build/lint if docs tooling arrives.
+1. go vet + go test ./... (root module; packages without ladybug cgo)
+2. `go test ./internal/brain/rank` (cgo-free ranking + flag parser)
+3. python -m unittest discover -s bin/tools (includes published-docs SoT)
+4. bin/facts/audit self  (lexicon internal consistency)
+5. bin/brain/eval.go       (recall@5 ≥ 0.95, gates index regressions)
+6. md-docs build/lint if docs tooling arrives.

 Feedback loop: every commit → PR → CI → green/gate → merge. Same discipline as
 `db/tech-poc`: contract first where there is an OpenAPI/message shape.
@@ -30,9 +30,9 @@ graph TB
    subgraph dph["2dph tools"]
        EX["bin/facts/extract<br/>2-source pairing"]
        AU["bin/facts/audit<br/>confidence + staleness"]
-        IDX["bin/kb/index<br/>chunk + embed"]
-        MD["bin/md/import<br/>mistune leaves"]
-        SR["bin/kb/search<br/>deduction + --hop"]
+        IDX["bin/brain/index.go<br/>chunk + embed"]
+        MD["bin/markdown/import.go<br/>mistune leaves"]
+        SR["bin/brain/search.go<br/>deduction"]
    end

    subgraph store["Ladybug var/kb.lbug"]
@@ -85,19 +85,33 @@ fact; conflicting sources or a single source → `hypothesis` → `(not confirme
 ## Deduction search

 ```bash
-bin/kb/search "Matrix federation over HTTPS"      # facts → info → web-search
-bin/kb/search "what runs on arc-2" --hop 1        # walk graph edges
-bin/kb/search "where is cs-lexicon" --json | yq '.'  # YAML by default
-bin/kb/get <id> --body                            # full chunk on demand
-bin/kb/stats                                      # index health
-bin/kb/eval                                       # recall@5 gate
+bin/brain/search.go "Matrix federation over HTTPS"   # facts → info → web-search
+bin/brain/search.go "onlyoffice postgres" --root facts
+bin/brain/search.go "where is cs-lexicon" --json | yq '.'
+bin/brain/get.go <id> --body                         # full chunk on demand
+bin/brain/stats.go                                   # index health
+bin/brain/eval.go                                    # recall@5 gate
+```
+
+`--hop` is not implemented (needs File/FROM_FILE edges); the flag errors instead of walking. `bin/kb/search` is a deprecated wrapper around `bin/brain/search.go`.
+
+Mail is a first-class corpus (retrievable through the same search):
+
+```bash
+bin/mail/sync.go --source onlyoffice,gmail --workers 8 --out var/mail  # raw sync (Go)
+bin/mail/import.go --from-raw var/mail                                  # JSON → markdown
+bin/brain/index.go --rebuild                                            # rebuild brain (incl. mail)
+bin/brain/search.go "invoice from last week"                            # same search over mail leafs
 ```

 ## Storage

 - **LadybugDB** — single `var/kb.lbug`, Cypher property graph, HNSW + BM25
  in one engine, embedded (no server), ACID, read-only-safe for concurrent
-  readers.
+  readers. **Never `DROP INDEX` FTS/VECTOR** on Ladybug 0.19: DROP leaves
+  ghost catalog tables (`_0_Leaf_vec_UPPER`) so recreate fails while
+  `SHOW_INDEXES` omits HNSW. Fresh indexes = delete `var/kb.lbug` +
+  `bin/brain/index.go --rebuild`. Use `ensure_indexes()` after upserts.
 - **model2vec** — `potion-multilingual-128M` static embeddings (256-dim),
  CPU-fast, deterministic, no Ollama runtime dependency.
 - facts and info split semantically by `root` column but written inside the
@@ -105,10 +119,10 @@ bin/kb/eval                                       # recall@5 gate

 ## Tooling conventions

-`bin/{subject}/{method}` — self-describing: shebang on line 1, usage comment
-from line 2. bash + python primary; golang via the Go shebang when a compiled
-helper is right. YAML default output, `--json` for machines. Everything that
-touches network/db is read-only, throttled, cached. Tests gate every commit.
+`bin/{subject}/{method}.go` — self-describing: shebang on line 1, usage comment
+from line 2. Shared code in `internal/`. YAML default output, `--json` for
+machines. Tests gate every commit. HTTP: `bin/brain/serve.go` calls
+`internal/brain` in-process (`/health` `/search` `/get` `/stats` `/audit` `/ingest`).

 ## Development

@@ -116,7 +130,7 @@ touches network/db is read-only, throttled, cached. Tests gate every commit.
 uv venv .venv                                  # Python 3.12, uv-managed
 uv pip install -r requirements.lock.txt        # pinned toolchain
 bin/facts/audit self                           # lexicon consistency gate
-go test ./... && python -m unittest discover -s tools -t .
+go test ./... && python -m unittest discover -s bin/tools -t .
 ```

 Docker (optional, cached model + var volumes):
@@ -124,7 +138,7 @@ Docker (optional, cached model + var volumes):
 ```bash
 docker compose run --rm brain index            # (re)index corpus
 docker compose run --rm brain search "query"   # one-shot query
-docker compose run --rm brain serve            # async Go HTTP server
+docker compose run --rm brain serve            # bin/brain/serve.go
 docker compose up brain-watch                  # auto re-index on change
 ```

@@ -134,6 +148,9 @@ docker compose up brain-watch                  # auto re-index on change
  Neo4j + Qdrant + Matrix RAG brain
 - [agent-skills](https://github.com/eSlider/agent-skills) — upstream
  skills (`web-search`, `db-yaml`, …) that 2dph integrates
- [detective](https://github.com/detective) — the two-source method
+- detective method — the two-source method

-See [PLAN.md](PLAN.md) for decisions, execution status, and v2 open questions.
+Work board (issues): [git.produktor.io/eSlider/2dph/issues](https://git.produktor.io/eSlider/2dph/issues).
+PRs and CI: GitHub [`eSlider/2dph`](https://github.com/eSlider/2dph).
+
+See [PLAN.md](PLAN.md) for decisions, execution status, and v2 open questions.
@@ -0,0 +1,3 @@
+// Commands in this directory are shebang mains (search.go, serve.go, index.go,
+// get.go, stats.go, eval.go, watch.go), each behind an exclusive build tag.
+package main
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=brain_eval "$0" "$@"; exit
+//go:build brain_eval
+//
+// bin/brain/eval.go - recall@5 gate.
+//
+//	./bin/brain/eval.go
+//	./bin/brain/eval.go --json
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/kb/eval", os.Args[1:]))
+}
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=brain_get "$0" "$@"; exit
+//go:build brain_get
+//
+// bin/brain/get.go - read one leaf by id.
+//
+//	./bin/brain/get.go <id>
+//	./bin/brain/get.go <id> --body
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/kb/get", os.Args[1:]))
+}
@@ -0,0 +1,24 @@
+//usr/bin/env go run -tags=brain_index "$0" "$@"; exit
+//go:build brain_index
+//
+// bin/brain/index.go - rebuild the Ladybug graph (Python write path).
+//
+//	./bin/brain/index.go --rebuild
+//	./bin/brain/index.go --rebuild --with-mail
+//	./bin/brain/index.go --dry-run --with-mail
+//
+// v1 write is always a rebuild when mail is included (live FTS/HNSW + bulk
+// insert corrupts Ladybug 0.19 WAL). `add` is v2.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	args := append([]string{"--with-mail"}, os.Args[1:]...)
+	os.Exit(cmdbin.ExecFile("bin/kb/index", args))
+}
@@ -0,0 +1,23 @@
+//usr/bin/env go run -tags=system_ladybug "$0" "$@"; exit
+//go:build cgo && system_ladybug
+//
+// bin/brain/search.go - deduction search over the 2dph brain.
+//
+//	./bin/brain/search.go "query" [--root facts|info] [--repo P] [-n N] [--json]
+//	./bin/brain/search.go serve [port]
+//	./bin/brain/search.go --list-model
+//
+// Needs CGO + libladybug (CGO_CFLAGS/CGO_LDFLAGS). Prefer the wrapper
+// bin/kb/search which sets those and builds a binary for the embed daemon.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/brain"
+)
+
+func main() {
+	os.Exit(brain.Main(os.Args[1:]))
+}
@@ -0,0 +1,31 @@
+//usr/bin/env go run -tags=brain_serve,system_ladybug "$0" "$@"; exit
+//go:build brain_serve && cgo && system_ladybug
+//
+// bin/brain/serve.go - HTTP API (in-process ladybug search).
+//
+//	KB_ROOT=/path/to/2dph ./bin/brain/serve.go
+//	KB_WORKERS=4 KB_PORT=8630 ./bin/brain/serve.go
+//
+// Needs CGO + libladybug (same as bin/brain/search.go).
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"log"
+	"os"
+
+	"github.com/eSlider/2dph/internal/brain"
+	"github.com/eSlider/2dph/internal/httpapi"
+)
+
+func main() {
+	if os.Getenv("KB_ROOT") == "" {
+		if wd, err := os.Getwd(); err == nil {
+			os.Setenv("KB_ROOT", wd)
+		}
+	}
+	if err := brain.Ready(); err != nil {
+		log.Fatal(err)
+	}
+	httpapi.Run(brain.HTTP{})
+}
@@ -0,0 +1,20 @@
+//go:build brain_serve && !system_ladybug
+//
+// Fallback serve when ladybug cgo is not in the build (CI / tags=brain_serve).
+// Production shebang is serve.go (in-process).
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/httpapi"
+)
+
+func main() {
+	if os.Getenv("KB_ROOT") == "" {
+		if wd, err := os.Getwd(); err == nil {
+			os.Setenv("KB_ROOT", wd)
+		}
+	}
+	httpapi.Run(nil)
+}
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=brain_stats "$0" "$@"; exit
+//go:build brain_stats
+//
+// bin/brain/stats.go - index health.
+//
+//	./bin/brain/stats.go
+//	./bin/brain/stats.go --json
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/kb/stats", os.Args[1:]))
+}
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=brain_watch "$0" "$@"; exit
+//go:build brain_watch
+//
+// bin/brain/watch.go - re-index when corpus files change.
+//
+//	./bin/brain/watch.go [dir...]
+//	KB_WATCH_INTERVAL=15 ./bin/brain/watch.go
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/bin/watch"
+)
+
+func main() {
+	watch.Run(os.Args[1:])
+}
@@ -0,0 +1,29 @@
+#!/usr/bin/env bash
+# bin/chats - sync, import, index, facts, apply for Telegram/WhatsApp/LinkedIn.
+# Builds the chats binary on first run / when source changes, then execs it.
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+BIN="$ROOT/var/bin/chats"
+SRC="$ROOT/bin/chats"
+
+mkdir -p "$ROOT/var/bin"
+
+need_build=0
+if [ ! -x "$BIN" ]; then
+    need_build=1
+else
+    while IFS= read -r -d '' f; do
+        if [ "$f" -nt "$BIN" ]; then
+            need_build=1
+            break
+        fi
+    done < <(find "$SRC" -name '*.go' -print0 2>/dev/null)
+fi
+
+if [ "$need_build" -eq 1 ]; then
+    echo "Building chats..." >&2
+    (cd "$SRC" && go build -o "$BIN" .) || exit 1
+fi
+
+exec "$BIN" "$@"
@@ -0,0 +1,19 @@
+//usr/bin/env go run -tags=chats_apply "$0" "$@"; exit
+//go:build chats_apply
+//
+// bin/chats/apply.go - push extracted chat facts to OnlyOffice CRM.
+//
+//	./bin/chats/apply.go [--dry-run]
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/chats"
+)
+
+func main() {
+	os.Exit(chats.RunApply(os.Args[1:]))
+}
@@ -0,0 +1,4 @@
+// Commands in this directory are shebang mains (sync.go, import.go, facts.go,
+// apply.go), each behind an exclusive build tag so `go build ./bin/chats`
+// does not see two mains. Shared code lives in internal/chats.
+package main
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=chats_facts "$0" "$@"; exit
+//go:build chats_facts
+//
+// bin/chats/facts.go - extract phone/email/linkedin facts from JSONL.
+//
+//	./bin/chats/facts.go
+//
+// Writes var/chats/facts/. Does not index the brain.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/chats"
+)
+
+func main() {
+	os.Exit(chats.RunFacts(os.Args[1:]))
+}
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=chats_import "$0" "$@"; exit
+//go:build chats_import
+//
+// bin/chats/import.go - JSONL → markdown under var/chats/md/.
+//
+//	./bin/chats/import.go
+//
+// Conversion only. Brain ingest is bin/brain/index.go, not this command.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/chats"
+)
+
+func main() {
+	os.Exit(chats.RunImport(os.Args[1:]))
+}
@@ -0,0 +1,155 @@
+#!/usr/bin/env python3
+"""chats/refresh-linkedin-session - refresh LinkedIn MCP session from webtop CDP.
+
+  bin/chats/refresh-linkedin-session [--cdp URL] [--root DIR]
+"""
+
+Reads the current LinkedIn cookies out of the running Thorium browser in the
+work-webtop container via CDP (Network.getAllCookies), copies the live browser
+profile onto the source profile directory, and rewrites the portable
+cookies.json + source-state.json that mcp-server-linkedin requires.
+
+Usage:
+  refresh-linkedin-session [--cdp http://127.0.0.1:9222] [--root /var/tmp/liprofile]
+                           [--container work-webtop] [--profile thorium-profile]
+
+After the headless driver uses a copied profile, LinkedIn rotates the session
+in that copy, so this must run before every sync.
+"""
+
+import asyncio
+import json
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+import urllib.request
+
+import websockets
+
+
+def cdp_tab(ws_json):
+    for t in ws_json:
+        if t.get("webSocketDebuggerUrl"):
+            return t["webSocketDebuggerUrl"]
+    return None
+
+
+async def get_cookies(ws_url):
+    async with websockets.connect(ws_url, max_size=50_000_000) as ws:
+        await ws.send(json.dumps({"id": 1, "method": "Network.getAllCookies", "params": {}}))
+        resp = await ws.recv()
+        return json.loads(resp).get("result", {}).get("cookies", [])
+
+
+def write_source_state(root, profile_dir):
+    # Reuse the linkedin-mcp-server session_state module to write a valid
+    # source-state.json (same schema the daemon reads).
+    try:
+        from linkedin_mcp_server.session_state import canonical, write_source_state
+
+        write_source_state(canonical(__import__("pathlib").Path(profile_dir)))
+        return
+    except Exception:
+        pass
+    # Fallback: minimal schema-compatible state.
+    import uuid
+
+    state = {
+        "version": 1,
+        "source_runtime_id": "linux-amd64-host",
+        "login_generation": str(uuid.uuid4()),
+        "created_at": None,
+        "profile_path": profile_dir,
+        "cookies_path": os.path.join(root, "cookies.json"),
+    }
+    from datetime import datetime, timezone
+
+    state["created_at"] = datetime.now(timezone.utc).isoformat()
+    with open(os.path.join(root, "source-state.json"), "w") as f:
+        json.dump(state, f, indent=2)
+
+
+def main():
+    args = sys.argv[1:]
+    cdp = "http://127.0.0.1:9222"
+    root = "/var/tmp/liprofile"
+    container = "work-webtop"
+    cprofile = "thorium-profile"
+    for i in range(0, len(args), 2):
+        k = args[i]
+        v = args[i + 1] if i + 1 < len(args) else ""
+        if k == "--cdp":
+            cdp = v
+        elif k == "--root":
+            root = v
+        elif k == "--container":
+            container = v
+        elif k == "--profile":
+            cprofile = v
+
+    profile_dir = os.path.join(root, "profile")
+    os.makedirs(profile_dir, exist_ok=True)
+
+    # 1. Clear stale daemon/browser locks so the server can claim the profile.
+    for lock in ("profile-claim.lock", "profile.lock", "daemon.lock", "lease.lock"):
+        p = os.path.join(root, lock)
+        if os.path.exists(p):
+            os.remove(p)
+    for name in os.listdir(profile_dir):
+        if name.startswith("Singleton"):
+            os.remove(os.path.join(profile_dir, name))
+    for name in os.listdir(root):
+        if name.startswith("invalid-state-"):
+            shutil.rmtree(os.path.join(root, name), ignore_errors=True)
+
+    # 1. Copy the live browser profile (cookies DB + Local State) so the
+    #    session the driver launches carries the current login.
+    subprocess.run(
+        ["docker", "cp", f"{container}:/config/{cprofile}/Default", os.path.join(profile_dir, "Default")],
+        check=True, capture_output=True,
+    )
+    subprocess.run(
+        ["docker", "cp", f"{container}:/config/{cprofile}/Local State", os.path.join(profile_dir, "Local State")],
+        check=True, capture_output=True,
+    )
+    for lock in ("SingletonLock", "SingletonCookie", "SingletonSocket"):
+        p = os.path.join(profile_dir, lock)
+        if os.path.exists(p):
+            os.remove(p)
+
+    # 2. Pull the live cookies out of the running browser.
+    with urllib.request.urlopen(f"{cdp}/json", timeout=5) as r:
+        tabs = json.loads(r.read())
+    ws_url = cdp_tab(tabs)
+    if not ws_url:
+        sys.stderr.write("refresh-linkedin-session: no CDP tab\n")
+        sys.exit(1)
+    cookies = asyncio.run(get_cookies(ws_url))
+
+    li = [c for c in cookies if "linkedin" in c.get("domain", "")]
+    out = []
+    for c in li:
+        domain = c.get("domain", "")
+        if domain in (".www.linkedin.com", "www.linkedin.com"):
+            domain = ".linkedin.com"
+        out.append({
+            "name": c["name"],
+            "value": c["value"].strip('"'),
+            "domain": domain,
+            "path": c.get("path", "/"),
+            "expires": c.get("expires", -1),
+            "httpOnly": c.get("httpOnly", False),
+            "secure": c.get("secure", False),
+            "sameSite": c.get("sameSite", "None"),
+        })
+    with open(os.path.join(root, "cookies.json"), "w") as f:
+        json.dump(out, f, indent=2)
+
+    write_source_state(root, profile_dir)
+    sys.stderr.write(f"refresh-linkedin-session: {len(out)} cookies, profile refreshed\n")
+
+
+if __name__ == "__main__":
+    main()
@@ -0,0 +1,41 @@
+//usr/bin/env go run -tags=chats_sync "$0" "$@"; exit
+//go:build chats_sync
+//
+// bin/chats/sync.go - download chat messages to var/chats/<platform>/.
+//
+//	./bin/chats/sync.go telegram [--limit N] [--phone PHONE]
+//	./bin/chats/sync.go linkedin [--limit N] [--refresh]
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"fmt"
+	"os"
+
+	"github.com/eSlider/2dph/internal/chats"
+)
+
+func main() {
+	if len(os.Args) < 2 {
+		fmt.Fprintln(os.Stderr, `usage: bin/chats/sync.go telegram|linkedin [flags]`)
+		os.Exit(2)
+	}
+	platform := os.Args[1]
+	args := os.Args[2:]
+	switch platform {
+	case "telegram":
+		os.Exit(chats.RunSyncTelegram(args))
+	case "linkedin":
+		os.Exit(chats.RunSyncLinkedIn(args))
+	case "whatsapp":
+		fmt.Fprintln(os.Stderr, "chats: WhatsApp not implemented yet")
+		os.Exit(1)
+	case "help", "-h", "--help":
+		fmt.Fprintln(os.Stderr, `usage: bin/chats/sync.go telegram|linkedin [flags]`)
+		return
+	default:
+		fmt.Fprintf(os.Stderr, "chats: unknown platform %q\n", platform)
+		os.Exit(2)
+	}
+}
@@ -17,7 +17,7 @@ import subprocess
 import sys
 from pathlib import Path

-sys.path.insert(0, str(Path(__file__).resolve().parents[2] / "tools"))
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "tools"))

 from semver import bump_type, bump_version  # noqa: E402

@@ -34,11 +34,13 @@ case "${1:-}" in
    ;;
  "")
    [ -f "$HOME/.ssh/config" ] || { echo "db/ssh-tunnel: ~/.ssh/config missing" >&2; exit 1; }
-    if db/ssh-tunnel --check; then
+    if "$0" --check; then
      echo "tunnel already up on ${SRC}"
      exit 0
    fi
    ssh -f -N -M -S "$HOME/.ssh/2dph-tunnel.sock" \
+      -o StrictHostKeyChecking=accept-new \
+      -o BatchMode=yes \
      -L "${SRC}:${DST}" -p "$SSH_PORT" "${SSH_USER}@${SSH_HOST}" \
      && echo "tunnel up on ${SRC} (-> vm:${DST})"
    exit 0
@@ -0,0 +1,2 @@
+// Deprecated shebang mains at bin root (serve.go is tagged brain_serve).
+package main
@@ -2,10 +2,12 @@
 # bin/docker-entrypoint - run 2dph tools inside the container.
 #
 #   brain                 shell (default)
-#   brain search <q>      bin/kb/search
-#   brain index           bin/kb/index
-#   brain watch <dir>     watchdog re-indexer
-#   brain serve           async Go HTTP server (serve/)
+#   brain search <q>      bin/brain/search.go
+#   brain index           bin/kb/index --with-mail
+#   brain watch <dir>     compiled /app/bin/watch (bin/brain/watch.go)
+#   brain serve           compiled /app/bin/serve (bin/brain/serve.go)
+#   brain extract         bin/facts/extract (docker×compose pairing)
+#   brain audit           bin/facts/audit
 #
 # Usage comment starts at line 2 (self-describing convention).
 set -euo pipefail
@@ -16,8 +18,10 @@ shift || true
 case "$CMD" in
  shell) exec bash ;;
  search) exec "$KB_PY" /app/bin/kb/search "$@" ;;
-  index) exec "$KB_PY" /app/bin/kb/index "$@" ;;
-  watch) exec bash /app/bin/kb-watch "$@" ;;
-  serve) exec /app/serve/serve "$@" ;;
+  index) exec "$KB_PY" /app/bin/kb/index --with-mail "$@" ;;
+  watch) exec /app/bin/watch "$@" ;;
+  serve) exec /app/bin/serve "$@" ;;
+  extract) exec "$KB_PY" /app/bin/facts/extract "$@" ;;
+  audit) exec "$KB_PY" /app/bin/facts/audit "$@" ;;
  *) echo "unknown command: $CMD" >&2; exit 2 ;;
-esac
+esac
@@ -20,7 +20,7 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))


 def audit_db() -> list[str]:
@@ -0,0 +1,124 @@
+#!/usr/bin/env python3
+"""facts/crm - prove person->company and company->project associations.
+
+Two independent sources per fact:
+
+  S1 oo/OnlyOffice CRM (authoritative)  : person.company_id -> company,
+                                          project.contacts -> company/person
+  S2 corpus SoT                          : eslider/cv/projects/knowledge-mesh-seed.yaml
+                                          (orgs: employer/client/...  + projects)
+
+Only associations supported by BOTH sources are written as root=facts.
+Mismatches are reported (or, with --fix-crm, printed as oo CLI commands).
+
+Usage:
+  bin/facts/crm                    write proven facts (needs var/kb.lbug)
+  bin/facts/crm --dry-run          show proposed facts + mismatches only
+  bin/facts/crm --mismatches       show associations found in only one side
+"""
+from __future__ import annotations
+
+import json
+import os
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "bin" / "tools"))
+
+from kblib import upsert_leaf, connect, leaf_id  # noqa: E402
+
+MESH_ENV = os.environ.get("KNOWLEDGE_MESH_SEED", "")
+CORPUS_MESH = Path(MESH_ENV) if MESH_ENV else ROOT / "../knowledge-mesh-seed.yaml"
+
+
+def corpus_orgs(raw: str) -> dict[str, dict]:
+    """Delegate to tools.crmfacts.corpus_orgs (tested in tools/)."""
+    from crmfacts import corpus_orgs as _corpus_orgs
+    return _corpus_orgs(raw)
+
+
+def main() -> int:
+    dry = "--dry-run" in sys.argv
+    mism = "--mismatches" in sys.argv
+
+    mesh = CORPUS_MESH.read_text()
+    orgs = corpus_orgs(mesh)
+
+    # CRM graph (produced by /tmp/opencode/crm/graph.py -> /tmp/opencode/crm/graph.json)
+    graph = json.load(open("/tmp/opencode/crm/graph.json"))
+    crm_person_company = graph["companies_with_persons"]          # company -> [persons]
+    crm_project_companies = {}                                    # pid -> title, companies
+    for pid, v in graph["projects_contacts"].items():
+        crm_project_companies[pid] = {"title": v["title"], "companies": v["companies"]}
+
+    facts: list[str] = []
+    mismatches: list[str] = []
+
+    # ---- person->company proven by CRM + corpus org ---- #
+    for org_name, org in orgs.items():
+        token = org.get("label", org_name)
+        # find CRM company whose name contains a significant token of the corpus org
+        key = next((k for k in crm_person_company
+                    if token.split()[0].lower() in k.lower() or any(
+                        t.lower() in k.lower() for t in org.get("label", "").split(" / "))),
+                   None)
+        persons = crm_person_company.get(key, []) if key else []
+        if persons and org:
+            for p in persons:
+                facts.append(f"{p} is associated with {org.get('label')} "
+                             f"(role: {org.get('kind', '?')}, {org.get('period', '')})")
+        elif org and key and not persons:
+            mismatches.append(f"corpus org '{org_name}' ({org.get('label')}) has no CRM persons")
+        elif org and not key:
+            mismatches.append(f"corpus org '{org_name}' ({org.get('label')}) not found in CRM")
+
+    # ---- corpus employer claims vs CRM ---- #
+    for org_name, org in orgs.items():
+        if not org or not org.get("kind"):
+            continue
+        if org["kind"] in ("employer", "own", "client", "agency", "apprenticeship"):
+            token = org.get("label", org_name).split()[0]
+            if not any(token.lower() in k.lower() for k in crm_person_company):
+                mismatches.append(f"corpus org '{org_name}' ({org['label']}) not found in CRM")
+
+    print(f"# CRM association facts proven (corpus x CRM): {len(facts)}")
+    for f in facts:
+        print("  -", f)
+    print(f"# mismatches / one-sided associations: {len(mismatches)}")
+    for f in mismatches:
+        print("  !", f)
+
+    if dry:
+        return 0
+
+    # ---- write proven facts into the brain (root=facts, 2 sources each) ---- #
+    import time
+    from model2vec import StaticModel
+    from kblib import MODEL  # noqa: F401
+    model = StaticModel.from_pretrained(MODEL)
+    db, conn = connect(read_only=False)
+    try:
+        r = conn.execute("MATCH (l:Leaf) WHERE l.root='facts' RETURN count(*) AS n")
+        stats_before = r.get_all()[0][0]
+    except Exception:
+        stats_before = 0
+    rev = time.strftime("%Y%m%d-%H%M%S")
+    written = 0
+    for f in facts:
+        src = f"ooCRM x {CORPUS_MESH.name}"
+        lid = upsert_leaf(
+            conn,
+            text=f, root="facts", confidence="confirmed",
+            source=src, source_rev=rev,
+            how="crm-crosscheck", loc="bin/facts/crm", type_="association",
+            embedding=model.encode(f).tolist(),
+        )
+        written += 1
+    conn.close()
+    print(f"# wrote {written} facts into var/kb.lbug (facts was {stats_before})")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
@@ -21,7 +21,7 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))

 COMPOSE_FILES = [ROOT / "docker" / "compose.yaml", ROOT / "compose.yaml"]
 DOC_MARKERS = ["README.md", "PLAN.md", "AGENTS.md"]
@@ -176,8 +176,7 @@ def dedupe(facts: list[dict]) -> list[dict]:


 def write_facts(facts: list[dict]) -> None:
-    from kblib import connect, init_schema, upsert_leaf
-    from kblib import VAR
+    from kblib import VAR, connect, ensure_indexes, init_schema, upsert_leaf
    VAR.mkdir(exist_ok=True)
    db, conn = connect(VAR / "kb.lbug", read_only=False)
    init_schema(conn)
@@ -188,6 +187,8 @@ def write_facts(facts: list[dict]) -> None:
        upsert_leaf(conn, text=f["text"], root="facts", confidence="confirmed",
                    source=f["source"], source_rev=REPO, how=f["how"],
                    loc=f["loc"], type_="fact", embedding=emb)
+    # Upsert-with-index is safe; never DROP+recreate (ghost catalog kills HNSW).
+    ensure_indexes(conn)
    conn.close()
    db.close()

@@ -0,0 +1,154 @@
+#!/usr/bin/env python3
+"""git/import - import git history (commits, authors, files) into the brain.
+
+  bin/git/import [REPO]                 import all commits -> leafs + graph
+  bin/git/import --json                 emit import leafs as JSON, no write
+  bin/git/import --limit 100            cap commits processed
+  bin/git/import --since 2026-01-01     only recent commits
+  bin/git/import --root DIR             run per repo dir under DIR
+  bin/git/import --no-env               never read .env anywhere (default: true)
+
+Reads `git log --no-merges --name-only` from the repo, maps commits to
+`info` leafs (root=info, type=commit) and writes the version graph
+`File -[:HAS_VERSION]-> Commit -[:AUTHORED]-> Person` into var/kb.lbug.
+Idempotent: leaf MERGE by (source,text via leaf_id), graph MERGE by sha.
+"""
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "bin" / "tools"))
+
+from kblib import (  # noqa: E402
+    connect, ensure_indexes, init_schema, upsert_leaf,
+)
+from gitimport import commits_to_leafs, ensure_git_schema, index_commits, parse_log  # noqa: E402
+
+LOG_FMT = "--format=%x1e%H%x1f%an%x1f%ae%x1f%aI%x1f%s"
+
+
+def git_log(repo: Path, limit: int = 0, since: str = "") -> str:
+    cmd = ["git", "-C", str(repo), "log", "--no-merges", "--name-only", LOG_FMT]
+    if since:
+        cmd += ["--since", since]
+    if limit:
+        cmd += ["-n", str(limit)]
+    try:
+        out = subprocess.run(cmd, capture_output=True, text=True, timeout=120)
+    except (FileNotFoundError, subprocess.TimeoutExpired):
+        return ""
+    if out.returncode != 0:
+        print(f"git/import: {repo}: {out.stderr.strip()}", file=sys.stderr)
+        return ""
+    return out.stdout
+
+
+def repo_name(repo: Path) -> str:
+    try:
+        out = subprocess.run(
+            ["git", "-C", str(repo), "remote", "get-url", "origin"],
+            capture_output=True, text=True, timeout=20)
+        url = out.stdout.strip()
+        return url.rstrip("/").split("/")[-1].removesuffix(".git") if url else repo.name
+    except (FileNotFoundError, subprocess.TimeoutExpired):
+        return repo.name
+
+
+def embedder():
+    from model2vec import StaticModel
+    model = StaticModel.from_pretrained("minishlab/potion-multilingual-128M")
+    return lambda text: model.encode([text])[0].astype(float).tolist()
+
+
+def import_repo(conn, repo: Path, embed, limit: int, since: str,
+                no_write: bool = False) -> tuple[int, int]:
+    raw = git_log(repo, limit, since)
+    commits = parse_log(raw)
+    leafs = commits_to_leafs(commits, repo_name(repo))
+    if no_write:
+        return len(commits), 0
+    written = 0
+    for lf in leafs:
+        query = f"{lf['heading']}\n\n{lf['text']}"
+        emb = embed(lf["text"]) if lf["text"] else None
+        upsert_leaf(conn, text=query, root="info", confidence="confirmed",
+                    source=lf["source"], source_rev="git", how="git/import",
+                    loc=lf["source"], type_=lf.get("type", "commit"),
+                    embedding=emb)
+        written += 1
+    index_commits(conn, commits, repo_name(repo))
+    return len(commits), written
+
+
+def main(argv: list[str]) -> int:
+    import argparse
+    p = argparse.ArgumentParser(description="import git history into the brain")
+    p.add_argument("repo", nargs="?", default=None)
+    p.add_argument("--root", default=None, help="directory of repos to import (each git dir separately)")
+    p.add_argument("--limit", type=int, default=0)
+    p.add_argument("--since", default="")
+    p.add_argument("--json", action="store_true")
+    p.add_argument("--dry-run", action="store_true", help="parse + report, no db write")
+    a = p.parse_args(argv)
+
+    repos: list[Path] = []
+    if a.repo:
+        repos = [Path(a.repo)]
+    elif a.root:
+        root = Path(a.root)
+        if root.is_file():
+            repos = [root]
+        else:
+            repos = [dp for dp in sorted(root.iterdir()) if (dp / ".git").exists() or dp.is_file()]
+    else:
+        repos = [ROOT]
+
+    total_commits = 0
+    results: list[dict] = []
+    if a.dry_run:
+        for repo in repos:
+            if not repo.exists():
+                continue
+            commits = parse_log(git_log(repo, a.limit, a.since))
+            name = repo_name(repo)
+            total_commits += len(commits)
+            results.append({"repo": name, "commits": len(commits),
+                            "leafs": len(commits_to_leafs(commits, name)), "path": str(repo)})
+        if a.json:
+            print(json.dumps(results, indent=2))
+        else:
+            for r in results:
+                print(f"{r['repo']:<24} {r['commits']:>5} commits -> {r['leafs']} leafs  {r['path']}")
+        return 0
+
+    # Never DROP FTS/VECTOR (ghost catalog). Upsert while indexes exist is OK;
+    # ensure_indexes only CREATEs when missing.
+    db, conn = connect(ROOT / "var" / "kb.lbug", read_only=False)
+    init_schema(conn)
+    embed = embedder()
+    rows: list[dict] = []
+    for repo in repos:
+        if not repo.exists():
+            continue
+        reached, written = import_repo(conn, repo, embed, a.limit, a.since)
+        total_commits += reached
+        rows.append({"repo": repo_name(repo), "commits": reached, "written": written})
+    ensure_indexes(conn)
+    conn.close()
+    db.close()
+
+    if a.json:
+        print(json.dumps(rows, indent=2))
+    else:
+        for r in rows:
+            print(f"imported {r['commits']:>5} commits -> {r['written']} leafs  {r['repo']}")
+        print(f"total: {total_commits} commits")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -1,27 +0,0 @@
-#!/usr/bin/env bash
-# kb-watch - re-index 2dph when corpus files change.
-#
-#   kb-watch [dir...] [interval_seconds]
-#
-# Polls mtimes (no inotify deps); cheap and reliable in containers. Defaults:
-# dirs = /corpus (compose) or . ; interval = 30s.
-set -euo pipefail
-
-DEFAULT_DIRS="${KB_WATCH_DIRS:-/corpus}"
-DIRS=("$@")
-[[ ${#DIRS[@]} -eq 0 ]] && DIRS=(${DEFAULT_DIRS})
-INTERVAL="${KB_WATCH_INTERVAL:-30}"
-
-index() { "${KB_PY:-python3}" /app/bin/kb/index; }
-
-LAST_STAMP=""
-while true; do
-  STAMP=$(find "${DIRS[@]}" -type f -newermt "-${INTERVAL} seconds" 2>/dev/null \
-    | head -1 | md5sum)
-  if [[ -n "$STAMP" && "$STAMP" != "$LAST_STAMP" ]]; then
-    echo "kb-watch: changes detected, re-indexing" >&2
-    index || echo "kb-watch: index failed; will retry" >&2
-    LAST_STAMP="$STAMP"
-  fi
-  sleep "$INTERVAL"
-done
@@ -13,7 +13,7 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))

 from kblib import open_readonly, query_fts  # noqa: E402
 from yamlout import to_yaml  # noqa: E402
@@ -10,7 +10,7 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))

 from kblib import open_readonly  # noqa: E402
 from yamlout import to_yaml  # noqa: E402
@@ -19,13 +19,14 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))

 from kblib import (  # noqa: E402
-    connect, create_fts_and_vector, init_schema, upsert_leaf,
+    connect, ensure_indexes, init_schema, upsert_leaf,
    open_readonly, stats,
 )
 from mdleaves import read_markdown, to_all, walk_markdown  # noqa: E402
+from mailleafs import from_mail_root  # noqa: E402

 CORPUS_DEFAULTS = ["README.md", "PLAN.md", "AGENTS.md", "docs", "skills"]

@@ -99,43 +100,60 @@ def main(argv: list[str]) -> int:
    p = argparse.ArgumentParser(description="build the 2dph brain index")
    p.add_argument("--corpus", action="append", help="extra markdown dir/file to index (may repeat)")
    p.add_argument("--rebuild", action="store_true", help="fresh db + indexes")
+    p.add_argument("--with-mail", action="store_true", help="include var/mail message.md leafs")
+    p.add_argument("--since", default="", help="with --with-mail, only messages dated >= YYYY-MM-DD")
+    p.add_argument("--dry-run", action="store_true", help="count leafs, write nothing")
+    p.add_argument(
+        "--skip-indexes",
+        action="store_true",
+        help="write leafs only; caller runs ensure_indexes after seeding facts",
+    )
    p.add_argument("--limit", type=int, default=0, help="max leafs to embed")
    p.add_argument("--json", action="store_true")
    a = p.parse_args(argv)

    from kblib import DB_PATH, VAR
-    VAR.mkdir(exist_ok=True)
-    if a.rebuild and DB_PATH.exists():
-        DB_PATH.unlink()

    leafs = load_corpus(ROOT)
    if a.corpus:
        for source in a.corpus:
            leafs.extend(load_corpus_glob(source))
+    mail_n = 0
+    if a.with_mail:
+        mail = from_mail_root(ROOT / "var" / "mail", since=a.since)
+        mail_n = len(mail)
+        leafs.extend(mail)
+
+    if a.dry_run:
+        msg = {"indexed": 0, "corpus_total": len(leafs), "mail_leafs": mail_n, "dry_run": True}
+        print(json.dumps(msg, indent=2) if a.json else
+              f"brain/index: {len(leafs)} leafs would be indexed (mail={mail_n})")
+        return 0
+
+    VAR.mkdir(exist_ok=True)
+    if a.rebuild and DB_PATH.exists():
+        DB_PATH.unlink()

    db, conn = connect(DB_PATH, read_only=False)
    init_schema(conn)

-    if not (a.rebuild or _already_indexed(conn)):
-        create_fts_and_vector(conn, force=True)
+    # Never DROP FTS/VECTOR (ghost catalog). Write leafs, then ensure indexes
+    # unless --skip-indexes (seed facts first — MERGE under live FTS corrupts it).
+    # --rebuild already deleted kb.lbug above, so CREATE runs on a clean DB.
    embed = embedder()
    done, total = index_leafs(conn, leafs, embed, a.limit)
-    create_fts_and_vector(conn, force=(done > 0 or a.rebuild))
+    if not a.skip_indexes:
+        ensure_indexes(conn)
    s = stats(conn)
    conn.close()
    db.close()

    result = {"indexed": done, "corpus_total": total, **{k: v for k, v in s.items() if k in ("total", "by_root")}}
+    if a.skip_indexes:
+        result["indexes"] = "skipped"
    print(json.dumps(result, indent=2) if a.json else f"indexed {done}/{total} leafs; db total {s['total']}")
    return 0


-def _already_indexed(conn) -> bool:
-    try:
-        return conn.execute("MATCH (l:Leaf) RETURN count(*)").get_all()[0][0] > 0
-    except Exception:
-        return False
-
-
 if __name__ == "__main__":
    sys.exit(main(sys.argv[1:]))
@@ -1,72 +1,37 @@
-#!/usr/bin/env python3
-"""kb/search - deduction search over the 2dph brain.
+#!/usr/bin/env bash
+# bin/kb/search — deprecated wrapper. Use bin/brain/search.go.
+# Sets CGO for ladybug, builds a binary (embed daemon needs a real executable),
+# then execs it. Prints one deprecation line.
+set -euo pipefail

-  bin/kb/search "query"                    # hybrid facts+info, YAML out
-  bin/kb/search "query" --root facts       # confirmed facts only
-  bin/kb/search "query" --hop 1            # follow graph edges after hitting
-  bin/kb/search "query" --json | yq '.'
-  bin/kb/search "query" -n 5               # more results
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
+BIN="$ROOT/var/bin/brain-search"
+SRC="$ROOT/internal/brain"
+CMD="$ROOT/bin/brain"

-Deduction order: facts root first (confirmed answers with evidence links),
-then info root (marked `(not confirmed)`). --root restricts to one root.
--hop N walks FROM_FILE edges (sibling leafs in the same source file).
-"""
-from __future__ import annotations
+mkdir -p "$ROOT/var/bin"

-import json
-import sys
-from pathlib import Path
+need_build=0
+if [ ! -x "$BIN" ]; then
+	need_build=1
+else
+	while IFS= read -r -d '' f; do
+		if [ "$f" -nt "$BIN" ]; then
+			need_build=1
+			break
+		fi
+	done < <(find "$SRC" "$CMD" -name '*.go' -print0 2>/dev/null)
+fi

-ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+if [ "$need_build" -eq 1 ]; then
+	echo "Building brain/search..." >&2
+	(
+		cd "$ROOT" &&
+			CGO_CFLAGS="-I$ROOT/lib-ladybug" \
+			CGO_LDFLAGS="-L$ROOT/lib-ladybug -Wl,-rpath,$ROOT/lib-ladybug" \
+			go build -tags system_ladybug -o "$BIN" ./bin/brain
+	) || exit 1
+fi

-from kblib import connect, hybrid_search, init_schema, open_readonly, query_fts  # noqa: E402
-from yamlout import to_yaml  # noqa: E402
-import ladybug  # noqa: E402
-
-
-def main(argv: list[str]) -> int:
-    import argparse
-    p = argparse.ArgumentParser(description="deduction search over the brain")
-    p.add_argument("query")
-    p.add_argument("--root", choices=("facts", "info", None), default=None)
-    p.add_argument("--hop", type=int, default=0)
-    p.add_argument("-n", "--limit", type=int, default=10)
-    p.add_argument("--json", action="store_true")
-    a = p.parse_args(argv)
-
-    try:
-        db, conn = open_readonly()
-    except FileNotFoundError as e:
-        print(e, file=sys.stderr)
-        return 1
-
-    from model2vec import StaticModel
-    model = StaticModel.from_pretrained("minishlab/potion-multilingual-128M")
-    emb = model.encode([a.query])[0].astype(float).tolist()
-
-    rhs: list[dict] = []
-    try:
-        rhs = query_fts(conn, a.query, a.limit * 2)
-    except Exception:
-        rhs = []
-
-    results = hybrid_search(conn, emb, rhs, a.limit)
-    if a.root:
-        results = [h for h in results if h["root"] == a.root]
-
-    for hit in results:
-        hit.pop("rrf", None)
-        if hit.get("text"):
-            hit["snippet"] = hit["text"][:280]
-
-    out = {"query": a.query, "root_filter": a.root or "facts+info",
-           "count": len(results), "results": results}
-    print(json.dumps(out, indent=2, ensure_ascii=False) if a.json else to_yaml(out))
-    conn.close()
-    db.close()
-    return 0
-
-
-if __name__ == "__main__":
-    sys.exit(main(sys.argv[1:]))
+echo "bin/kb/search is deprecated; use bin/brain/search.go" >&2
+exec "$BIN" "$@"
@@ -11,7 +11,7 @@ import sys
 from pathlib import Path

 ROOT = Path(__file__).resolve().parents[2]
-sys.path.insert(0, str(ROOT / "tools"))
+sys.path.insert(0, str(ROOT / "bin" / "tools"))

 from kblib import open_readonly, stats  # noqa: E402
 from yamlout import to_yaml  # noqa: E402
@@ -0,0 +1,23 @@
+//usr/bin/env go run "$0" "$@"; exit
+// bin/kb/watch.go — deprecated. Use bin/brain/watch.go.
+//
+// Usage:
+//
+//	./bin/kb/watch.go [dir...]            # dirs default /corpus
+//	KB_WATCH_INTERVAL=15 ./bin/kb/watch.go
+//
+// Shebang trick: first line is a Go `//` comment; the real code lives in the
+// importable package (module path, never a relative import).
+// NOTE: never run `gofmt -w` on this file - it rewrites `//usr/bin/env` to
+// `// usr/...` and breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/bin/watch"
+)
+
+func main() {
+	watch.Run(os.Args[1:])
+}
@@ -0,0 +1,450 @@
+#!/usr/bin/env python3
+"""mail/import - pull OnlyOffice mails into var/mail/ as markdown.
+
+  bin/mail/import --from-raw var/mail   convert Go-synced message.json to md
+  bin/mail/import                       import newest inbox messages
+  bin/mail/import --folder sent         import sent folder
+  bin/mail/import --since 2026-01-01    only messages after a date
+  bin/mail/import --limit 50            cap messages per run
+  bin/mail/import --no-attachments      body only, skip attachment conversion
+  bin/mail/import --ocr                OCR scanned PDFs/images via docling
+  bin/mail/import --dry-run            list messages without writing anything
+
+Writes one directory per message: var/mail/{folder}/{message_id}/
+  message.md                frontmatter + markdown body
+  attachments/              raw attachment files (zips unpacked to _unpacked/)
+  attachments/*.md          converted attachment content
+
+Indexing is a separate step (`bin/brain/index.go --rebuild`): conversion can
+crash in native docling and must not leave the brain DB mid-transaction.
+
+Requires ONLYOFFICE_URL/USER/PASS in .env (or env). Idempotent: a message
+already present (message.md exists) is skipped unless --force.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import re
+import subprocess
+import sys
+import time
+import urllib.parse
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+sys.path.insert(0, str(ROOT / "bin" / "tools"))
+
+from mailconv import (  # noqa: E402
+    ARCHIVE_SUFFIXES,
+    IMAGE_SUFFIXES,
+    LEGACY_OFFICE_SUFFIXES,
+    TEXT_SUFFIXES,
+    html_to_markdown,
+    is_convertible,
+    normalize_markdown,
+    subject_to_filename,
+    zip_extract_safe,
+)
+
+import requests  # noqa: E402
+
+FOLDER_IDS = {"inbox": 1, "sent": 2, "drafts": 3, "trash": 4, "spam": 5}
+DEFAULT_LIMIT = 25
+
+
+def load_env() -> dict:
+    env = {k: v for k, v in os.environ.items()}
+    envfile = ROOT / ".env"
+    if envfile.exists():
+        for line in envfile.read_text().splitlines():
+            line = line.strip()
+            if not line or line.startswith("#") or "=" not in line:
+                continue
+            k, _, v = line.partition("=")
+            env.setdefault(k.strip(), v.strip().strip("\"'"))
+    url = env.get("ONLYOFFICE_URL") or env.get("OO_URL")
+    user = env.get("ONLYOFFICE_USER") or env.get("OO_USER")
+    password = env.get("ONLYOFFICE_PASS") or env.get("OO_PASSWORD")
+    missing = [n for n, v in (("ONLYOFFICE_URL", url), ("ONLYOFFICE_USER", user),
+                              ("ONLYOFFICE_PASS", password)) if not v]
+    if missing:
+        sys.exit(f"mail/import: missing {', '.join(missing)} (need .env or env)")
+    return {"url": url.rstrip("/"), "user": user, "password": password}
+
+
+class OOClient:
+    def __init__(self, conf: dict):
+        self.base = conf["url"]
+        self.session = requests.Session()
+        self.token = None
+        self._login(conf)
+
+    def _login(self, conf: dict) -> None:
+        r = self.session.post(f"{self.base}/api/2.0/authentication.json",
+                              json={"userName": conf["user"], "password": conf["password"], "type": 0},
+                              timeout=30)
+        r.raise_for_status()
+        body = r.json()
+        self.token = (body.get("response") or {}).get("token", "")
+        if not self.token:
+            sys.exit("mail/import: authentication failed (empty token)")
+
+    def _headers(self) -> dict:
+        return {"Authorization": f"Bearer {self.token}", "Accept": "application/json"}
+
+    def get(self, path: str, params: dict | None = None):
+        r = self.session.get(f"{self.base}{path}", params=params, headers=self._headers(), timeout=30)
+        r.raise_for_status()
+        return r.json()
+
+    def list_messages(self, folder: int, page: int = 1, count: int = DEFAULT_LIMIT) -> list[dict]:
+        data = self.get("/api/2.0/mail/messages",
+                        params={"folder": folder, "page": page, "count": count})
+        return data.get("response", [])
+
+    def get_message(self, message_id: str) -> dict:
+        data = self.get(f"/api/2.0/mail/messages/{message_id}")
+        return data.get("response", {})
+
+    def download_attachment(self, attach_id, dest: Path) -> bool:
+        """Download one attachment via the portal session cookie (.ashx handler)."""
+        url = f"{self.base}/addons/mail/httphandlers/download.ashx?attachid={attach_id}"
+        r = self.session.get(url, timeout=60)
+        if r.status_code != 200:
+            return False
+        dest.parent.mkdir(parents=True, exist_ok=True)
+        dest.write_bytes(r.content)
+        return True
+
+
+def folder_id(name: str) -> int:
+    if name in FOLDER_IDS:
+        return FOLDER_IDS[name]
+    if name.isdigit():
+        return int(name)
+    sys.exit(f"mail/import: unknown folder '{name}' (use {', '.join(FOLDER_IDS)})")
+
+
+def safe_attachment_name(att: dict) -> str:
+    name = att.get("fileName") or att.get("storedName") or "attachment"
+    name = re.sub(r"[^\w.\- ]+", "_", name)
+    return name
+
+
+def convert_file_to_md(path: Path, ocr: bool) -> str | None:
+    """Convert one attachment file to markdown text; None when not convertible."""
+    suffix = path.suffix.lower()
+    if suffix in TEXT_SUFFIXES:
+        return normalize_markdown(path.read_text(encoding="utf-8", errors="replace"))
+    if suffix in (".docx", ".pptx", ".xlsx", ".html", ".htm", ".epub", ".eml", ".msg"):
+        try:
+            from markitdown import MarkItDown
+            md = MarkItDown()
+            result = md.convert(str(path))
+            return normalize_markdown(result.text_content)
+        except Exception as e:
+            return f"\n<!-- conversion failed: {e} -->\n"
+    if suffix == ".pdf":
+        return _convert_pdf(path, ocr)
+    if suffix in IMAGE_SUFFIXES and ocr:
+        return _convert_pdf(path, ocr)
+    if suffix in LEGACY_OFFICE_SUFFIXES:
+        return _convert_legacy(path)
+    if suffix in ARCHIVE_SUFFIXES:
+        return None  # handled by caller (unpack + recurse)
+    return None
+
+
+def _convert_pdf(path: Path, ocr: bool) -> str:
+    """Convert one PDF to markdown.
+
+    Fast path: poppler's pdftotext (-layout) extracts exact text from
+    born-digital PDFs in ~15ms vs docling's 1-3s. Only textless PDFs (scanned
+    pages, layout-heavy) fall back to docling, which runs isolated in a
+    subprocess because its native onnx/RT-DETR has segfaulted the main process.
+    """
+    text = _pdf_fast_text(path)
+    if ocr or text is None or not text.strip():
+        return _convert_pdf_docling(path, ocr)
+    return normalize_markdown(text)
+
+
+def _pdf_fast_text(path: Path) -> str | None:
+    """pdftotext -layout; None when poppler is unavailable (or the PDF has no text layer)."""
+    try:
+        proc = subprocess.run(
+            ["pdftotext", "-layout", str(path), "-"],
+            capture_output=True, timeout=60)
+    except (OSError, subprocess.TimeoutExpired):
+        return None
+    if proc.returncode != 0:
+        return None
+    return proc.stdout.decode("utf-8", errors="replace")
+
+
+def _convert_pdf_docling(path: Path, ocr: bool) -> str:
+    try:
+        proc = subprocess.run(
+            [sys.executable, os.path.abspath(__file__), "--pdf-worker", str(path),
+             "--ocr" if ocr else "--no-ocr"],
+            capture_output=True, text=True, timeout=600)
+    except subprocess.TimeoutExpired:
+        return "\n<!-- pdf conversion timed out -->\n"
+    if proc.returncode != 0:
+        tail = proc.stderr.strip().splitlines()[-3:]
+        return f"\n<!-- pdf conversion failed: {proc.returncode}: {' | '.join(tail)} -->\n"
+    return proc.stdout
+
+
+def _pdf_worker(path: Path, ocr: bool) -> None:
+    """docling worker entry: prints converted markdown on stdout, exits non-zero on error."""
+    try:
+        from docling.document_converter import DocumentConverter, PdfFormatOption
+        from docling.datamodel.pipeline_options import PdfPipelineOptions
+        opts = PdfPipelineOptions()
+        opts.do_ocr = bool(ocr)
+        opts.do_table_structure = True
+        conv = DocumentConverter(format_options={"pdf": PdfFormatOption(pipeline_options=opts)})
+        res = conv.convert(str(path))
+        sys.stdout.write(normalize_markdown(res.document.export_to_markdown()))
+        sys.exit(0)
+    except Exception as e:
+        # errors/stacktraces to stderr; the caller only reports a one-liner
+        print(f"pdf-worker: {e}", file=sys.stderr)
+        import traceback
+        traceback.print_exc(file=sys.stderr)
+        sys.exit(1)
+
+
+def _convert_legacy(path: Path) -> str:
+    """Legacy .doc/.xls/.ppt -> md via pandoc (installed) or a stub."""
+    try:
+        out = subprocess.run(["pandoc", str(path), "-t", "markdown"],
+                             capture_output=True, text=True, timeout=120)
+        if out.returncode == 0 and out.stdout.strip():
+            return normalize_markdown(out.stdout)
+    except (FileNotFoundError, subprocess.TimeoutExpired):
+        pass
+    return f"\n<!-- legacy {path.suffix} not convertible (pandoc unavailable) -->\n"
+
+
+def write_message_md(msg: dict, folder: str, out_dir: Path, target_dir: Path | None = None) -> Path:
+    import yaml
+    body_html = msg.get("htmlBody") or ""
+    body_text = msg.get("textBody") or ""
+    body_md = ""
+    if body_html.strip():
+        body_md = html_to_markdown(body_html)
+    elif body_text.strip():
+        body_md = normalize_markdown(body_text)
+    # accept both OnlyOffice (receivedDate) and Go-sync (receivedAt) date keys
+    date = msg.get("receivedDate") or msg.get("receivedAt") or ""
+    if date and not isinstance(date, str):
+        date = str(date)
+    meta = {
+        "id": msg.get("id"),
+        "source": msg.get("source"),
+        "folder": folder,
+        "subject": msg.get("subject", ""),
+        "from": msg.get("from", ""),
+        "to": msg.get("to", ""),
+        "cc": msg.get("cc", ""),
+        "date": date,
+        "has_attachments": bool(msg.get("hasAttachments")),
+        "mime_message_id": msg.get("mimeMessageId", ""),
+        "calendar_uid": msg.get("calendarUid", ""),
+        "type": "mail",
+    }
+    meta = {k: v for k, v in meta.items() if v not in (None, "")}
+    frontmatter = "---\n" + yaml.safe_dump(meta, sort_keys=False, allow_unicode=True).strip() + "\n---\n"
+    content = f"{frontmatter}\n# {meta.get('subject','')}\n\n{body_md}".strip() + "\n"
+    if target_dir is not None:
+        msg_dir = target_dir
+    else:
+        msg_dir = out_dir / folder / str(meta.get("id"))
+    msg_dir.mkdir(parents=True, exist_ok=True)
+    md_path = msg_dir / "message.md"
+    md_path.write_text(content, encoding="utf-8")
+    return md_path
+
+
+def convert_attachments(msg: dict, msg_dir: Path, ocr: bool) -> list[dict]:
+    """Download + convert each attachment; returns [{name, md, raw}] summaries.
+
+    raw file keeps the API storedName (unique hash, avoids collisions); the
+    markdown is named after the friendly fileName when available.
+
+    In --from-raw mode attachments are already on disk (Go sync wrote them;
+    .ics already has a structured .md sidecar). Files with an existing .md
+    sidecar are left as-is, only unconverted raws are converted here.
+    """
+    out: list[dict] = []
+    atts = msg.get("attachments") or []
+    att_dir = msg_dir / "attachments"
+    for att in atts:
+        aid = att.get("fileId")
+        display = safe_attachment_name(att)
+        stored = att.get("storedName")
+        raw_name = safe_attachment_name({"storedName": stored}) if stored else display
+        raw = att_dir / raw_name
+        if aid and not raw.exists() and OOCLIENT is not None:
+            if not OOCLIENT.download_attachment(aid, raw):
+                out.append({"name": display, "md": "\n<!-- download failed -->\n", "raw": str(raw)})
+                continue
+        if not raw.exists():
+            out.append({"name": display, "md": "\n<!-- raw missing -->\n", "raw": str(raw)})
+            continue
+        md_stem = Path(display).stem or raw.stem
+        md_path = att_dir / f"{md_stem}.md"
+        # Go sync pre-wrote structured .md for .ics; keep it.
+        if not md_path.exists():
+            md_text = _convert_att_recursive(raw, ocr)
+            md_path.write_text(f"# Attachment: {display}\n\n{md_text}\n", encoding="utf-8")
+        else:
+            md_text = md_path.read_text(encoding="utf-8", errors="replace")
+        out.append({"name": display, "md": md_text, "raw": str(raw), "md_file": str(md_path)})
+    return out
+
+
+def _convert_att_recursive(path: Path, ocr: bool) -> str:
+    if path.suffix.lower() in ARCHIVE_SUFFIXES:
+        parts: list[str] = []
+        unpack = path.parent / "_unpacked" / path.stem
+        files = zip_extract_safe(path, unpack)
+        for f in files:
+            sub = _convert_att_recursive(f, ocr)
+            if sub and sub.strip():
+                parts.append(f"## {f.name}\n\n{sub}")
+        return "\n\n".join(parts) if parts else "\n<!-- empty zip -->\n"
+    text = convert_file_to_md(path, ocr)
+    return text or "\n<!-- not convertible -->\n"
+
+
+# module-level client for attachment downloads in convert_attachments
+OOCLIENT: OOClient | None = None
+
+
+def convert_one(msg_dir: Path, full: dict, folder: str, out_root: Path,
+                ocr: bool, no_attachments: bool, target_dir: Path | None = None) -> dict:
+    """Write message.md + convert attachments for one message dict.
+
+    Works for both live API messages and the Go-sync message.json shape
+    (source field optional; attachments read from attachments/ dir).
+    target_dir overrides the derived path (used by --from-raw where the
+    directory layout is authoritative, not the message folder field).
+    """
+    mid = str(full.get("id"))
+    write_message_md(full, folder, out_root, target_dir=target_dir)
+    converted: list[dict] = []
+    if not no_attachments:
+        converted = convert_attachments(full, msg_dir, ocr)
+    return {"id": mid, "subject": full.get("subject", ""),
+            "date": full.get("receivedDate", "") or full.get("receivedAt", ""),
+            "attachments": len(converted)}
+
+
+def main(argv: list[str]) -> int:
+    global OOCLIENT
+    p = argparse.ArgumentParser(description="pull OnlyOffice mails to var/mail as markdown")
+    p.add_argument("--folder", default="inbox", help="inbox|sent|drafts|trash|spam or numeric id")
+    p.add_argument("--limit", type=int, default=DEFAULT_LIMIT, help="max messages per run")
+    p.add_argument("--offset", type=int, default=0, help="skip N messages")
+    p.add_argument("--since", default="", help="only messages received after YYYY-MM-DD")
+    p.add_argument("--id", action="append", default=[], help="import specific message id (repeatable)")
+    p.add_argument("--from-raw", default="",
+                   help="convert Go-synced dirs (var/mail/<folder>/<id>/message.json) to markdown")
+    p.add_argument("--no-attachments", action="store_true", help="skip attachment download+convert")
+    p.add_argument("--ocr", action="store_true", help="OCR scanned PDFs/images via docling")
+    p.add_argument("--force", action="store_true", help="re-import even if message.md exists")
+    p.add_argument("--dry-run", action="store_true", help="list messages, write nothing")
+    p.add_argument("--json", action="store_true")
+    p.add_argument("--pdf-worker", default="", help=argparse.SUPPRESS)
+    p.add_argument("--no-ocr", action="store_true", help=argparse.SUPPRESS)
+    a = p.parse_args(argv)
+
+    if a.pdf_worker:
+        _pdf_worker(Path(a.pdf_worker), ocr=not a.no_ocr)
+        return 0
+
+    conf = load_env()
+    fid = folder_id(a.folder)
+    out_root = ROOT / "var" / "mail"
+    summary: list[dict] = []
+    if a.from_raw:
+        OOCLIENT = None
+        raw_root = Path(a.from_raw)
+        for msg_dir in sorted(raw_root.rglob("message.json")):
+            mid = msg_dir.parent.name
+            entry = {"id": mid, "subject": "", "date": "",
+                     "attachments": 0, "skipped": False}
+            md_path = msg_dir.parent / "message.md"
+            if md_path.exists() and not a.force:
+                entry["skipped"] = True
+                summary.append(entry)
+                continue
+            if a.dry_run:
+                entry["skipped"] = "dry-run"
+                summary.append(entry)
+                continue
+            full = json.loads(msg_dir.read_text(encoding="utf-8"))
+            entry.update(convert_one(msg_dir.parent, full, full.get("folder") or a.folder,
+                                     raw_root, a.ocr, a.no_attachments,
+                                     target_dir=msg_dir.parent))
+            summary.append(entry)
+    else:
+        OOCLIENT = OOClient(conf)
+        if a.id:
+            messages = [{"id": i} for i in a.id]
+        else:
+            page = 1
+            messages = []
+            want = a.offset + a.limit
+            while len(messages) < want:
+                count = min(DEFAULT_LIMIT, want - len(messages))
+                chunk = OOCLIENT.list_messages(fid, page=page, count=count)
+                if not chunk:
+                    break
+                messages.extend(chunk)
+                if len(chunk) < count:
+                    break
+                page += 1
+            messages = messages[a.offset:a.offset + a.limit]
+            if a.since:
+                messages = [m for m in messages
+                            if (m.get("receivedDate") or "") >= a.since]
+
+        for m in messages:
+            mid = str(m.get("id"))
+            entry = {"id": mid, "subject": m.get("subject", ""),
+                     "date": m.get("receivedDate", ""), "attachments": 0, "skipped": False}
+            msg_dir = out_root / a.folder / mid
+            md_path = msg_dir / "message.md"
+            if md_path.exists() and not a.force:
+                entry["skipped"] = True
+                summary.append(entry)
+                continue
+            if a.dry_run:
+                entry["skipped"] = "dry-run"
+                summary.append(entry)
+                continue
+            full = OOCLIENT.get_message(mid)
+            entry.update(convert_one(msg_dir, full, a.folder, out_root, a.ocr, a.no_attachments,
+                                     target_dir=msg_dir))
+            summary.append(entry)
+    if a.json:
+        print(json.dumps(summary, ensure_ascii=False, indent=2))
+    else:
+        imported = [e for e in summary if not e["skipped"]]
+        print(f"mail/import: folder={a.folder} checked={len(summary)} "
+              f"imported={len(imported)} (skipped={sum(e['skipped'] is True for e in summary)})")
+        for e in summary:
+            flag = "skip" if e["skipped"] is True else ("dry" if e["skipped"] == "dry-run" else "ok  ")
+            print(f"  [{flag}] {e['id']} {e['date'][:10]} {e['subject'][:60]}"
+                  f" (atts={e['attachments']})")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=mail_import "$0" "$@"; exit
+//go:build mail_import
+//
+// bin/mail/import.go - message.json → markdown (no brain write).
+//
+//	./bin/mail/import.go --from-raw var/mail
+//
+// Indexing is bin/brain/index.go --rebuild, not this command.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/mail/import", os.Args[1:]))
+}
@@ -0,0 +1,27 @@
+#!/usr/bin/env python3
+"""mail/index_mail — deprecated. Use bin/brain/index.go --rebuild --with-mail.
+
+Ladybug corrupts its WAL on bulk-insert into an already-indexed DB, so this
+shim always rebuilds (repo corpus + var/mail). Conversion stays in mail/import.
+"""
+from __future__ import annotations
+
+import os
+import sys
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+def main(argv: list[str]) -> int:
+    print(
+        "bin/mail/index_mail is deprecated; use bin/brain/index.go --rebuild --with-mail",
+        file=sys.stderr,
+    )
+    index = ROOT / "bin" / "kb" / "index"
+    os.execv(sys.executable, [sys.executable, str(index), "--rebuild", "--with-mail", *argv])
+    return 1
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
@@ -0,0 +1,25 @@
+//usr/bin/env go run "$0" "$@"; exit
+// bin/mail/sync.go - async download of OnlyOffice and Gmail mail to var/mail/.
+//
+//	./bin/mail/sync.go --source onlyoffice,gmail --limit 50 --workers 8
+//	./bin/mail/sync.go --source gmail --force
+//	./bin/mail/sync.go --dry-run
+//
+// Writes raw message.json + attachments under var/mail/<folder>/<id>/; run
+// bin/mail/import.go --from-raw afterwards to convert everything to markdown.
+//
+// Shebang trick: first line is a Go `//` comment; the real code lives in the
+// importable package (module path, never a relative import).
+// NOTE: never run `gofmt -w` on this file - it rewrites `//usr/bin/env` to
+// `// usr/...` and breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/bin/mail/sync"
+)
+
+func main() {
+	os.Exit(sync.Main(os.Args[1:]))
+}
@@ -0,0 +1,156 @@
+// Package synccmd wires the sync library to a CLI: reads .env, parses flags,
+// picks sources, prints stats. Kept separate from the library so unit tests
+// don't depend on os.Args/env.
+package sync
+
+import (
+	"context"
+	"flag"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+)
+
+// CLIConfig is a superset of SyncConfig plus flag parsing results.
+type CLIConfig struct {
+	Sync SyncConfig
+	Env  string // .env path; default <cwd>/.env
+	Sources string
+	Help bool
+}
+
+// ParseCLI reads os.Args into a CLIConfig. Exit codes: 0 ok, 2 usage.
+func ParseCLI(args []string) (CLIConfig, int, error) {
+	fs := flag.NewFlagSet("mail/sync", flag.ContinueOnError)
+	var (
+		env    = fs.String("env", "", ".env file (default: <cwd>/.env)")
+		out    = fs.String("out", "", "var/mail root (default: <cwd>/var/mail)")
+		workers = fs.Int("workers", 4, "concurrent downloads")
+		limit  = fs.Int("limit", 0, "max messages per source (0 = all)")
+		offset = fs.Int("offset", 0, "skip first N messages per source")
+		force  = fs.Bool("force", false, "overwrite existing message.json + attachments")
+		dryRun = fs.Bool("dry-run", false, "list message counts without writing")
+		query  = fs.String("query", "in:inbox", "Gmail search query (gmail source only)")
+		srcs   = fs.String("source", "onlyoffice", "comma list: onlyoffice,gmail (default onlyoffice)")
+		help   = fs.Bool("help", false, "usage")
+	)
+	fs.SetOutput(os.Stderr)
+	if err := fs.Parse(args); err != nil {
+		return CLIConfig{}, 2, err
+	}
+	if *help || fs.NArg() > 0 {
+		return CLIConfig{Help: true}, 0, nil
+	}
+	wd, err := os.Getwd()
+	if err != nil {
+		return CLIConfig{}, 2, err
+	}
+	if *env == "" {
+		*env = filepath.Join(wd, ".env")
+	}
+	if *out == "" {
+		*out = filepath.Join(wd, "var", "mail")
+	}
+	envVars := readEnv(*env)
+	cfg := SyncConfig{
+		Out:     *out,
+		Workers: *workers,
+		Limit:   *limit,
+		Offset:  *offset,
+		Force:   *force,
+		DryRun:  *dryRun,
+		Query:   *query,
+		Policy:  RetryPolicy{},
+	}
+	cli := CLIConfig{Sync: cfg, Env: *env, Sources: *srcs}
+	for _, s := range strings.Split(*srcs, ",") {
+		switch strings.TrimSpace(s) {
+		case "onlyoffice":
+			u := pick(envVars["ONLYOFFICE_URL"], envVars["OO_URL"])
+			user := pick(envVars["ONLYOFFICE_USER"], envVars["OO_USER"])
+			pass := pick(envVars["ONLYOFFICE_PASS"], envVars["OO_PASSWORD"])
+			if u == "" || user == "" || pass == "" {
+				return CLIConfig{}, 2, fmt.Errorf("onlyoffice source needs ONLYOFFICE_URL/USER/PASS in %s", *env)
+			}
+			cfg.OO = &OOConfig{URL: u, User: user, Password: pass}
+		case "gmail":
+			home, _ := os.UserHomeDir()
+			cfg.Gmail = &GmailCredentials{
+				CredentialsPath: filepath.Join(home, ".gmail-mcp", "credentials.json"),
+				KeysPath:        filepath.Join(home, ".gmail-mcp", "gcp-oauth.keys.json"),
+			}
+		default:
+			return CLIConfig{}, 2, fmt.Errorf("unknown source %q", s)
+		}
+	}
+	cli.Sync = cfg
+	return cli, 0, nil
+}
+
+// Main is the CLI entry: returns process exit code.
+func Main(args []string) int {
+	cli, code, err := ParseCLI(args)
+	if err != nil {
+		fmt.Fprintln(os.Stderr, "mail/sync:", err)
+		return code
+	}
+	if cli.Help {
+		fmt.Fprintln(os.Stderr, "usage: bin/mail/sync.go [--source onlyoffice,gmail] [--query GMAIL_Q] [--limit N] [--offset N] [--workers N] [--force] [--dry-run]")
+		return 0
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 6*time.Hour)
+	defer cancel()
+	start := time.Now()
+	stats, err := Run(ctx, cli.Sync)
+	if err != nil {
+		fmt.Fprintln(os.Stderr, "mail/sync:", err)
+		return 1
+	}
+	if cli.Sync.DryRun {
+		fmt.Printf("mail/sync: dry-run checked=%d (no writes)\n", stats.Checked)
+		return 0
+	}
+	fmt.Printf("mail/sync: checked=%d new=%d skipped=%d failed=%d in %s\n",
+		stats.Checked, stats.New, stats.Skipped, stats.Failed, time.Since(start).Round(time.Millisecond))
+	if stats.Failed > 0 {
+		return 1
+	}
+	return 0
+}
+
+// readEnv parses KEY=VALUE lines (ignoring comments) with KEY=PATH override.
+func readEnv(path string) map[string]string {
+	out := map[string]string{}
+	b, err := os.ReadFile(path)
+	if err != nil {
+		return out
+	}
+	for _, line := range strings.Split(string(b), "\n") {
+		line = strings.TrimSpace(line)
+		if line == "" || strings.HasPrefix(line, "#") || !strings.Contains(line, "=") {
+			continue
+		}
+		k, v, _ := strings.Cut(line, "=")
+		out[strings.TrimSpace(k)] = strings.Trim(strings.TrimSpace(v), "\"'")
+	}
+	// env overrides file
+	for _, kv := range os.Environ() {
+		k, v, ok := strings.Cut(kv, "=")
+		if !ok {
+			continue
+		}
+		if strings.HasPrefix(k, "ONLYOFFICE_") || strings.HasPrefix(k, "OO_") {
+			out[k] = v
+		}
+	}
+	return out
+}
+
+func pick(a, b string) string {
+	if a != "" {
+		return a
+	}
+	return b
+}
@@ -0,0 +1,356 @@
+package sync
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"net/url"
+	"os"
+	"path/filepath"
+	"strings"
+	"time"
+)
+
+// GmailCredentials holds the OAuth files produced by the gmail MCP
+// (@gongrzhe/server-gmail-autoauth-mcp) auto-auth flow.
+type GmailCredentials struct {
+	CredentialsPath string // ~/.gmail-mcp/credentials.json
+	KeysPath        string // ~/.gmail-mcp/gcp-oauth.keys.json
+	User            string // fixed: the authed account
+}
+
+// gmailToken is the JSON shape of credentials.json + refresh response.
+type gmailToken struct {
+	AccessToken  string `json:"access_token"`
+	RefreshToken string `json:"refresh_token"`
+	Expiry       int64  `json:"expiry_date"` // ms epoch
+}
+
+type gmailKeys struct {
+	Installed *gmailKeyBlock `json:"installed"`
+	Web       *gmailKeyBlock `json:"web"`
+}
+type gmailKeyBlock struct {
+	ClientID     string `json:"client_id"`
+	ClientSecret string `json:"client_secret"`
+}
+
+// GmailClient talks to the Gmail REST API using the OAuth refresh token from
+// ~/.gmail-mcp/. Token is refreshed lazily with a mutex-guarded cache.
+type GmailClient struct {
+	creds  GmailCredentials
+	client *http.Client
+	mu     chan struct{}
+	token  *gmailToken
+	user   string
+}
+
+func NewGmailClient(creds GmailCredentials) (*GmailClient, error) {
+	if creds.CredentialsPath == "" {
+		home, _ := os.UserHomeDir()
+		creds.CredentialsPath = filepath.Join(home, ".gmail-mcp", "credentials.json")
+		creds.KeysPath = filepath.Join(home, ".gmail-mcp", "gcp-oauth.keys.json")
+	}
+	g := &GmailClient{
+		creds:  creds,
+		client: &http.Client{Timeout: 60 * time.Second},
+		mu:     make(chan struct{}, 1),
+	}
+	g.mu <- struct{}{}
+	return g, nil
+}
+
+// accessToken returns a fresh bearer token, refreshing via the Google token
+// endpoint when the cached one is missing or about to expire.
+func (g *GmailClient) accessToken(ctx context.Context) (string, error) {
+	select {
+	case <-g.mu:
+	case <-ctx.Done():
+		return "", ctx.Err()
+	}
+	defer func() { g.mu <- struct{}{} }()
+	if g.token != nil && g.token.AccessToken != "" && g.token.Expiry > time.Now().UnixMilli()+300_000 {
+		return g.token.AccessToken, nil
+	}
+	return g.refreshLocked(ctx)
+}
+
+func (g *GmailClient) refreshLocked(ctx context.Context) (string, error) {
+	cred, err := os.ReadFile(g.creds.CredentialsPath)
+	if err != nil {
+		return "", fmt.Errorf("read gmail credentials %s: %w", g.creds.CredentialsPath, err)
+	}
+	var t gmailToken
+	if err := json.Unmarshal(cred, &t); err != nil {
+		return "", fmt.Errorf("parse gmail credentials: %w", err)
+	}
+	if t.RefreshToken == "" {
+		return "", errors.New("gmail credentials.json has no refresh_token (run the gmail MCP auth flow)")
+	}
+	keys, err := os.ReadFile(g.creds.KeysPath)
+	if err != nil {
+		return "", fmt.Errorf("read gmail keys %s: %w", g.creds.KeysPath, err)
+	}
+	var k gmailKeys
+	if err := json.Unmarshal(keys, &k); err != nil {
+		return "", fmt.Errorf("parse gmail keys: %w", err)
+	}
+	block := k.Installed
+	if block == nil {
+		block = k.Web
+	}
+	if block == nil {
+		return "", errors.New("gmail gcp-oauth.keys.json has no installed/web block")
+	}
+
+	form := url.Values{}
+	form.Set("client_id", block.ClientID)
+	form.Set("client_secret", block.ClientSecret)
+	form.Set("refresh_token", t.RefreshToken)
+	form.Set("grant_type", "refresh_token")
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, "https://oauth2.googleapis.com/token",
+		strings.NewReader(form.Encode()))
+	if err != nil {
+		return "", err
+	}
+	req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	resp, err := g.client.Do(req)
+	if err != nil {
+		return "", fmt.Errorf("gmail token refresh: %w", err)
+	}
+	defer resp.Body.Close()
+	body, _ := io.ReadAll(io.LimitReader(resp.Body, 1<<20))
+	if resp.StatusCode != http.StatusOK {
+		var e struct {
+			Error string `json:"error"`
+			Desc  string `json:"error_description"`
+		}
+		_ = json.Unmarshal(body, &e)
+		if e.Error == "invalid_grant" {
+			return "", fmt.Errorf("gmail OAuth token invalid/expired - re-auth via: npx -y @gongrzhe/server-gmail-autoauth-mcp auth (uses ~/.gmail-mcp)")
+		}
+		return "", fmt.Errorf("gmail token refresh status %d: %s", resp.StatusCode, truncate(string(body), 300))
+	}
+	var out struct {
+		AccessToken string `json:"access_token"`
+		ExpiresIn   int64  `json:"expires_in"`
+	}
+	if err := json.Unmarshal(body, &out); err != nil {
+		return "", fmt.Errorf("gmail token refresh parse: %w", err)
+	}
+	g.token = &gmailToken{
+		AccessToken:  out.AccessToken,
+		RefreshToken: t.RefreshToken,
+		Expiry:       time.Now().UnixMilli() + out.ExpiresIn*1000,
+	}
+	return out.AccessToken, nil
+}
+
+// ListIDs returns message ids matching q, walking nextPageToken up to maxIDs
+// (0 = unlimited). Thread-level pagination via the messages.list endpoint.
+func (g *GmailClient) ListIDs(ctx context.Context, q string, maxIDs int, pageToken string) (ids []string, next string, err error) {
+	for {
+		params := url.Values{}
+		params.Set("q", q)
+		params.Set("maxResults", "100")
+		if pageToken != "" {
+			params.Set("pageToken", pageToken)
+		}
+		var out struct {
+			Messages []struct {
+				ID string `json:"id"`
+			} `json:"messages"`
+			NextPageToken string `json:"nextPageToken"`
+		}
+		if err := g.getJSON(ctx, "/gmail/v1/users/me/messages?"+params.Encode(), &out); err != nil {
+			return nil, "", err
+		}
+		for _, m := range out.Messages {
+			ids = append(ids, m.ID)
+			if maxIDs > 0 && len(ids) >= maxIDs {
+				return ids, out.NextPageToken, nil
+			}
+		}
+		if out.NextPageToken == "" {
+			break
+		}
+		pageToken = out.NextPageToken
+	}
+	return ids, "", nil
+}
+
+// GetMessage fetches a message in format=full and normalizes it.
+func (g *GmailClient) GetMessage(ctx context.Context, id string) (*Message, error) {
+	var raw struct {
+		ID        string `json:"id"`
+		ThreadID  string `json:"threadId"`
+		InternalDate string `json:"internalDate"` // ms epoch string
+		Payload   gmailPart
+	}
+	path := "/gmail/v1/users/me/messages/" + url.PathEscape(id) + "?format=full"
+	if err := g.getJSON(ctx, path, &raw); err != nil {
+		return nil, err
+	}
+	m := &Message{
+		Source: "gmail",
+		ID:     raw.ID,
+		Folder: "gmail",
+	}
+	for _, h := range raw.Payload.Headers {
+		switch strings.ToLower(h.Name) {
+		case "subject":
+			m.Subject = h.Value
+		case "from":
+			m.From = h.Value
+		case "to":
+			m.To = h.Value
+		case "cc":
+			m.CC = h.Value
+		case "bcc":
+			m.BCC = h.Value
+		case "message-id":
+			m.MimeMessageID = h.Value
+		case "date":
+			if t, err := time.Parse(time.RFC1123Z, h.Value); err == nil {
+				m.ReceivedAt = t
+			}
+		}
+	}
+	if ms, err := parseMS(raw.InternalDate); err == nil {
+		m.ReceivedAt = ms
+	}
+	m.TextBody, m.HTMLBody, m.Attachments = collectParts(raw.Payload, "root", m.ID, 0)
+	m.HasAttachments = len(m.Attachments) > 0
+	return m, nil
+}
+
+type gmailPart struct {
+	PartID   string      `json:"partId"`
+	MimeType string      `json:"mimeType"`
+	Filename string      `json:"filename"`
+	Body     gmailBody   `json:"body"`
+	Headers  []gmailHeader `json:"headers"`
+	Parts    []gmailPart `json:"parts"`
+}
+type gmailHeader struct {
+	Name  string `json:"name"`
+	Value string `json:"value"`
+}
+type gmailBody struct {
+	Size         int64  `json:"size"`
+	Data         string `json:"data"`
+	AttachmentID string `json:"attachmentId"`
+}
+
+// collectParts walks the MIME tree: text bodies into plain/html, anything with
+// a filename into attachments (returned with base64 ids for later download).
+func collectParts(p gmailPart, mime string, msgID string, depth int) (text, html string, atts []Attachment) {
+	if depth > 16 {
+		return
+	}
+	mt := strings.ToLower(p.MimeType)
+	if p.Filename != "" && mt != "text/plain" && mt != "text/html" {
+		pid := p.PartID
+		if pid == "" {
+			pid = fmt.Sprintf("%d", depth)
+		}
+		// Gmail's attachments API keys off body.attachmentId, not partId.
+		attID := p.Body.AttachmentID
+		if attID == "" {
+			attID = pid
+		}
+		atts = append(atts, Attachment{
+			FileID:      msgID + ":" + attID,
+			FileName:    p.Filename,
+			StoredName:  p.Filename,
+			Size:        p.Body.Size,
+			ContentType: p.MimeType,
+		})
+	} else if data, err := base64.URLEncoding.DecodeString(p.Body.Data); err == nil && len(p.Body.Data) > 0 {
+		s := string(data)
+		if mt == "text/html" && html == "" {
+			html = s
+		} else if (mt == "text/plain" || mt == "") && text == "" {
+			text = s
+		}
+	}
+	for _, child := range p.Parts {
+		t, h, a := collectParts(child, mt, msgID, depth+1)
+		if text == "" {
+			text = t
+		}
+		if html == "" {
+			html = h
+		}
+		atts = append(atts, a...)
+	}
+	return
+}
+
+// DownloadAttachment fetches an attachment's bytes from the Gmail API.
+func (g *GmailClient) DownloadAttachment(ctx context.Context, msgID, attID string) ([]byte, error) {
+	// attID format is "<msgId>:<partId>"; the API needs the bare attachment id.
+	partID := attID
+	if i := strings.Index(attID, ":"); i >= 0 {
+		partID = attID[i+1:]
+	}
+	var out struct {
+		Data string `json:"data"`
+	}
+	path := "/gmail/v1/users/me/messages/" + url.PathEscape(msgID) + "/attachments/" + url.PathEscape(partID)
+	if err := g.getJSON(ctx, path, &out); err != nil {
+		return nil, err
+	}
+	return base64.URLEncoding.DecodeString(out.Data)
+}
+
+func (g *GmailClient) getJSON(ctx context.Context, path string, out any) error {
+	tok, err := g.accessToken(ctx)
+	if err != nil {
+		return err
+	}
+	u := "https://gmail.googleapis.com" + path
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return err
+	}
+	req.Header.Set("Authorization", "Bearer "+tok)
+	resp, err := g.client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+	body, _ := io.ReadAll(io.LimitReader(resp.Body, 16<<20))
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("gmail %s: status %d: %s", path, resp.StatusCode, truncate(string(body), 300))
+	}
+	if out != nil {
+		return json.Unmarshal(body, out)
+	}
+	return nil
+}
+
+func parseMS(s string) (time.Time, error) {
+	if s == "" {
+		return time.Time{}, errors.New("empty")
+	}
+	var ms int64
+	if _, err := fmt.Sscanf(s, "%d", &ms); err != nil {
+		return time.Time{}, err
+	}
+	return time.UnixMilli(ms), nil
+}
+
+func truncate(s string, n int) string {
+	if len(s) <= n {
+		return s
+	}
+	return s[:n] + "…"
+}
+
+var _ = bytes.MinRead
@@ -0,0 +1,258 @@
+package sync
+
+import (
+	"fmt"
+	"strings"
+	"time"
+	"unicode/utf8"
+
+	ics "github.com/arran4/golang-ical"
+	"golang.org/x/text/encoding/charmap"
+)
+
+// ICSToMarkdown parses a VCALENDAR/VEVENT payload and renders a compact
+// structured markdown block: what / when / where / organizer / attendees.
+// Returns the raw text when the payload is not a calendar.
+func ICSToMarkdown(data []byte) string {
+	data = normalizeEncoding(data)
+	cal, err := ics.ParseCalendar(strings.NewReader(string(data)))
+	if err != nil {
+		return normalizeMarkdown(string(data))
+	}
+	method := ""
+	for _, p := range cal.CalendarProperties {
+		if p.IANAToken == string(ics.ComponentPropertyMethod) {
+			method = p.Value
+			break
+		}
+	}
+	method = strings.TrimSpace(method)
+	var out []string
+	for _, ev := range cal.Events() {
+		summary := strings.TrimSpace(propValue(ev, ics.ComponentPropertySummary))
+		if summary != "" {
+			out = append(out, "# "+summary)
+		}
+		if when := eventWhen(ev); when != "" {
+			out = append(out, "- **When:** "+when)
+		}
+		if loc := strings.TrimSpace(propValue(ev, ics.ComponentPropertyLocation)); loc != "" {
+			out = append(out, "- **Where:** "+loc)
+		}
+		if desc := strings.TrimSpace(stripHTML(propValue(ev, ics.ComponentPropertyDescription))); desc != "" {
+			out = append(out, "- **What:** "+desc)
+		}
+		if org := propValue(ev, ics.ComponentPropertyOrganizer); org != "" {
+			out = append(out, "- **Organizer:** "+attendeeFmt(org))
+		}
+		for _, a := range ev.Attendees() {
+			cn := strings.TrimSpace(firstParam(a.ICalParameters, "CN"))
+			partstat := string(a.ParticipationStatus())
+			name := cn
+			if name == "" {
+				name = a.Email()
+			}
+			line := name
+			if email := a.Email(); email != "" && email != name {
+				line = name + " <" + email + ">"
+			}
+			if partstat != "" && !strings.EqualFold(partstat, "NEEDS-ACTION") {
+				line += " (" + strings.Title(strings.ToLower(strings.ReplaceAll(partstat, "_", " "))) + ")"
+			}
+			out = append(out, "- **Attendee:** "+line)
+		}
+	}
+	if len(out) == 0 {
+		return normalizeMarkdown(string(data))
+	}
+	if method != "" {
+		out = append([]string{"*Calendar method: " + method + "*"}, out...)
+	}
+	return normalizeMarkdown(strings.Join(out, "\n\n"))
+}
+
+func eventWhen(ev *ics.VEvent) string {
+	start, errStart := ev.GetStartAt()
+	end, errEnd := ev.GetEndAt()
+	// All-day events: golang-ical has dedicated getters.
+	if errStart != nil {
+		if allDay, err := ev.GetAllDayStartAt(); err == nil {
+			start = allDay
+			errStart = nil
+		}
+	}
+	if errEnd != nil {
+		if allDay, err := ev.GetAllDayEndAt(); err == nil {
+			end = allDay
+			errEnd = nil
+		}
+	}
+	if errStart != nil {
+		// Non-IANA TZID (e.g. "W. Europe Standard Time"): parse the raw
+		// property text instead of failing.
+		return rawWhen(ev)
+	}
+	if errEnd != nil || end.Equal(start) {
+		return dtFmt(start)
+	}
+	return dtFmt(start) + " → " + dtFmt(end)
+}
+
+// rawWhen parses DTSTART/DTEND property values that golang-ical cannot resolve
+// because the TZID is not an IANA zone. Formats: 20260812T120000 or 20260812.
+func rawWhen(ev *ics.VEvent) string {
+	start := rawPropValue(ev, ics.ComponentPropertyDtStart)
+	end := rawPropValue(ev, ics.ComponentPropertyDtEnd)
+	if start == "" {
+		return ""
+	}
+	if end == "" || end == start {
+		return rawDTFmt(start)
+	}
+	return rawDTFmt(start) + " → " + rawDTFmt(end)
+}
+
+func rawPropValue(ev *ics.VEvent, prop ics.ComponentProperty) string {
+	p := ev.GetProperty(prop)
+	if p == nil {
+		return ""
+	}
+	return p.Value
+}
+
+// rawDTFmt turns 20260812T120000 into 2026-08-12 12:00; 20260812 into 2026-08-12.
+func rawDTFmt(s string) string {
+	s = strings.TrimSpace(s)
+	if len(s) >= 8 && isDigits(s[:8]) {
+		y, m, d := s[:4], s[4:6], s[6:8]
+		if len(s) > 8 && (s[8] == 'T' || s[8] == 't') && len(s) >= 15 && isDigits(s[9:15]) {
+			h, mi := s[9:11], s[11:13]
+			return fmt.Sprintf("%s-%s-%s %s:%s", y, m, d, h, mi)
+		}
+		return fmt.Sprintf("%s-%s-%s", y, m, d)
+	}
+	return s
+}
+
+func isDigits(s string) bool {
+	for _, c := range s {
+		if c < '0' || c > '9' {
+			return false
+		}
+	}
+	return s != ""
+}
+
+// dtFmt renders a time as local "2006-01-02 15:04" (tz label when meaningful).
+func dtFmt(t time.Time) string {
+	loc := t.Local()
+	label := ""
+	if loc.Location() != time.Local {
+		label = " " + loc.Location().String()
+	}
+	return loc.Format("2006-01-02 15:04") + label
+}
+
+// propertyGetter is satisfied by both *ics.Calendar and *ics.VEvent.
+type propertyGetter interface {
+	GetProperty(ics.ComponentProperty) *ics.IANAProperty
+}
+
+func propValue(ev propertyGetter, prop ics.ComponentProperty) string {
+	p := ev.GetProperty(prop)
+	if p == nil {
+		return ""
+	}
+	return p.Value
+}
+
+func firstParam(params map[string][]string, key string) string {
+	if vs, ok := params[key]; ok && len(vs) > 0 {
+		return vs[0]
+	}
+	return ""
+}
+
+func attendeeFmt(raw string) string {
+	raw = strings.TrimSpace(raw)
+	if i := strings.Index(raw, ":"); i >= 0 {
+		raw = raw[i+1:]
+	}
+	return raw
+}
+
+// stripHTML removes tags and decodes entities from an ics DESCRIPTION that may
+// carry HTML (Outlook/Exchange style), keeping text lines readable.
+func stripHTML(s string) string {
+	if !strings.Contains(s, "<") {
+		return s
+	}
+	lines := strings.Split(s, "\n")
+	for i, l := range lines {
+		var b strings.Builder
+		depth := 0
+		for j := 0; j < len(l); j++ {
+			c := l[j]
+			if c == '<' {
+				if j+1 < len(l) && l[j+1] == '/' {
+					depth--
+				} else {
+					depth++
+				}
+				for j < len(l) && l[j] != '>' {
+					j++
+				}
+				continue
+			}
+			if c == '>' {
+				continue
+			}
+			if depth == 0 {
+				b.WriteByte(c)
+			}
+		}
+		lines[i] = strings.TrimSpace(b.String())
+	}
+	return strings.Join(lines, "\n")
+}
+
+// normalizeMarkdown collapses blank-line runs and strips control chars.
+func normalizeMarkdown(s string) string {
+	s = strings.ReplaceAll(s, "\x00", "")
+	for _, ch := range []string{"\ufeff", "\u200b", "\u034f", "\u00ad", "\u2007", "\u2008", "\u200a", "\u2002"} {
+		s = strings.ReplaceAll(s, ch, "")
+	}
+	lines := strings.Split(s, "\n")
+	var out []string
+	blank := 0
+	for _, l := range lines {
+		if strings.TrimSpace(l) == "" {
+			blank++
+			if blank > 1 {
+				continue
+			}
+		} else {
+			blank = 0
+		}
+		out = append(out, l)
+	}
+	return strings.Join(out, "\n")
+}
+
+// normalizeEncoding re-encodes legacy single-byte text as UTF-8. ICS files
+// exported by some portals are Latin-1 (e.g. "N\xfcrnberg"); golang-ical
+// passes the bytes through, producing invalid UTF-8 in the markdown output.
+// Valid UTF-8 is returned untouched.
+func normalizeEncoding(data []byte) []byte {
+	if utf8.Valid(data) {
+		return data
+	}
+	dec := charmap.ISO8859_1.NewDecoder()
+	out, err := dec.Bytes(data)
+	if err != nil {
+		return data
+	}
+	return out
+}
+
+var _ = fmt.Sprintf // keep fmt import if helpers change
@@ -0,0 +1,222 @@
+package sync
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"net/http/cookiejar"
+	"net/url"
+	"strings"
+	"time"
+)
+
+// OOConfig mirrors the .env / environment used by bin/mail/import.
+type OOConfig struct {
+	URL      string
+	User     string
+	Password string
+}
+
+// OOClient is a minimal OnlyOffice API client: authentication.json for the
+// bearer token plus the session cookie jar required by the .ashx download
+// handler. It mirrors the endpoint contract bin/mail/import already uses.
+type OOClient struct {
+	cfg      OOConfig
+	client   *http.Client
+	mu       chan struct{}
+	token    string
+	folderID int
+}
+
+func NewOOClient(cfg OOConfig, folderID int) (*OOClient, error) {
+	jar, err := cookiejar.New(nil)
+	if err != nil {
+		return nil, err
+	}
+	c := &OOClient{
+		cfg:      cfg,
+		client:   &http.Client{Jar: jar, Timeout: 60 * time.Second},
+		mu:       make(chan struct{}, 1),
+		folderID: folderID,
+	}
+	c.mu <- struct{}{}
+	if err := c.authenticate(context.Background()); err != nil {
+		return nil, err
+	}
+	return c, nil
+}
+
+func (o *OOClient) authenticate(ctx context.Context) error {
+	select {
+	case <-o.mu:
+	case <-ctx.Done():
+		return ctx.Err()
+	}
+	defer func() { o.mu <- struct{}{} }()
+	body, _ := json.Marshal(map[string]any{
+		"userName": o.cfg.User, "password": o.cfg.Password, "type": 0,
+	})
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost,
+		strings.TrimRight(o.cfg.URL, "/")+"/api/2.0/authentication.json",
+		strings.NewReader(string(body)))
+	if err != nil {
+		return err
+	}
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := o.client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+	data, _ := io.ReadAll(io.LimitReader(resp.Body, 4<<20))
+	if resp.StatusCode < 200 || resp.StatusCode > 299 {
+		return fmt.Errorf("oo authenticate status %d: %s", resp.StatusCode, truncate(string(data), 200))
+	}
+	var out struct {
+		Response struct {
+			Token string `json:"token"`
+		} `json:"response"`
+	}
+	if err := json.Unmarshal(data, &out); err != nil {
+		return err
+	}
+	if out.Response.Token == "" {
+		return fmt.Errorf("oo authenticate: empty token")
+	}
+	o.token = out.Response.Token
+	return nil
+}
+
+// get performs an authenticated GET and decodes the JSON body into out.
+func (o *OOClient) get(ctx context.Context, path string, out any) error {
+	u := strings.TrimRight(o.cfg.URL, "/") + path
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return err
+	}
+	req.Header.Set("Authorization", "Bearer "+o.token)
+	req.Header.Set("Accept", "application/json")
+	resp, err := o.client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+	data, _ := io.ReadAll(io.LimitReader(resp.Body, 16<<20))
+	if resp.StatusCode < 200 || resp.StatusCode > 299 {
+		return fmt.Errorf("oo %s: status %d: %s", path, resp.StatusCode, truncate(string(data), 300))
+	}
+	if out != nil {
+		return json.Unmarshal(data, out)
+	}
+	return nil
+}
+
+// ooMessage mirrors the OnlyOffice mail message JSON (subset we need).
+type ooMessage struct {
+	ID             int    `json:"id"`
+	Subject        string `json:"subject"`
+	From           string `json:"from"`
+	To             string `json:"to"`
+	CC             string `json:"cc"`
+	BCC            string `json:"bcc"`
+	ReceivedDate   string `json:"receivedDate"`
+	HTMLBody       string `json:"htmlBody"`
+	TextBody       string `json:"textBody"`
+	HasAttachments bool   `json:"hasAttachments"`
+	MimeMessageID  string `json:"mimeMessageId"`
+	Attachments    []struct {
+		FileID      int    `json:"fileId"`
+		FileName    string `json:"fileName"`
+		StoredName  string `json:"storedName"`
+		Size        int64  `json:"size"`
+		ContentType string `json:"contentType"`
+	} `json:"attachments"`
+}
+
+// ListIDs returns message ids in the configured folder, paginating pages until
+// maxIDs is reached (0 = all).
+func (o *OOClient) ListIDs(ctx context.Context, maxIDs int, page int) (ids []int, next int, err error) {
+	var out struct {
+		Response []ooMessage `json:"response"`
+	}
+	count := 100
+	if maxIDs > 0 && maxIDs < count {
+		count = maxIDs
+	}
+	path := fmt.Sprintf("/api/2.0/mail/messages?folder=%d&page=%d&count=%d", o.folderID, page, count)
+	if err := o.get(ctx, path, &out); err != nil {
+		return nil, 0, err
+	}
+	for _, m := range out.Response {
+		ids = append(ids, m.ID)
+		if maxIDs > 0 && len(ids) >= maxIDs {
+			break
+		}
+	}
+	next = page + 1
+	return ids, next, nil
+}
+
+// GetMessage fetches the full message by id and normalizes into Message.
+func (o *OOClient) GetMessage(ctx context.Context, id int) (*Message, error) {
+	var out struct {
+		Response ooMessage `json:"response"`
+	}
+	path := fmt.Sprintf("/api/2.0/mail/messages/%d", id)
+	if err := o.get(ctx, path, &out); err != nil {
+		return nil, err
+	}
+	m := out.Response
+	msg := &Message{
+		Source:         "onlyoffice",
+		ID:             fmt.Sprintf("%d", m.ID),
+		Folder:         "oo",
+		Subject:        m.Subject,
+		From:           m.From,
+		To:             m.To,
+		CC:             m.CC,
+		BCC:            m.BCC,
+		HTMLBody:       m.HTMLBody,
+		TextBody:       m.TextBody,
+		HasAttachments: m.HasAttachments,
+		MimeMessageID:  m.MimeMessageID,
+	}
+	if t, err := time.Parse(time.RFC3339Nano, m.ReceivedDate); err == nil {
+		msg.ReceivedAt = t
+	}
+	for _, a := range m.Attachments {
+		msg.Attachments = append(msg.Attachments, Attachment{
+			FileID:      fmt.Sprintf("%d", a.FileID),
+			FileName:    a.FileName,
+			StoredName:  a.StoredName,
+			Size:        a.Size,
+			ContentType: a.ContentType,
+		})
+	}
+	return msg, nil
+}
+
+// DownloadAttachment fetches attachment bytes via the .ashx handler, which
+// requires the session cookie (client.Jar) captured during authenticate().
+func (o *OOClient) DownloadAttachment(ctx context.Context, fileID string) ([]byte, error) {
+	u := strings.TrimRight(o.cfg.URL, "/") + "/addons/mail/httphandlers/download.ashx?attachid=" + url.QueryEscape(fileID)
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return nil, err
+	}
+	resp, err := o.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	data, err := io.ReadAll(io.LimitReader(resp.Body, 64<<20))
+	if err != nil {
+		return nil, err
+	}
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("oo download attach %s: status %d", fileID, resp.StatusCode)
+	}
+	return data, nil
+}
@@ -0,0 +1,397 @@
+package sync
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"math"
+	"math/rand"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"sync/atomic"
+	"time"
+)
+
+// RetryPolicy is the exponential-backoff strategy applied to transient HTTP
+// failures (5xx, timeouts, network errors). Callers wrap transient errors with
+// retryWrap; everything else aborts immediately.
+type RetryPolicy struct {
+	MaxAttempts int           // total attempts (>=1); 0 => 5
+	BaseDelay   time.Duration // first backoff; 0 => 250ms
+	MaxDelay    time.Duration // cap; 0 => 15s
+	Jitter      float64       // 0..1 multiplier; 0 => 0.2
+}
+
+func (p RetryPolicy) withDefaults() RetryPolicy {
+	if p.MaxAttempts <= 0 {
+		p.MaxAttempts = 5
+	}
+	if p.BaseDelay <= 0 {
+		p.BaseDelay = 250 * time.Millisecond
+	}
+	if p.MaxDelay <= 0 {
+		p.MaxDelay = 15 * time.Second
+	}
+	if p.Jitter <= 0 {
+		p.Jitter = 0.2
+	}
+	return p
+}
+
+// delay returns the wait before attempt n (1-based): base * 2^(n-2) + jitter,
+// capped at MaxDelay. Attempt 1 waits 0, attempt 2 waits base, then doubles.
+func (p RetryPolicy) delay(attempt int) time.Duration {
+	if attempt <= 1 {
+		return 0
+	}
+	exp := math.Min(float64(attempt-2), 10)
+	d := float64(p.BaseDelay) * math.Pow(2, exp)
+	if p.Jitter > 0 {
+		d *= 1 - p.Jitter + 2*p.Jitter*rand.Float64()
+	}
+	if d > float64(p.MaxDelay) {
+		d = float64(p.MaxDelay)
+	}
+	return time.Duration(d)
+}
+
+type errRetry struct{ err error }
+
+func (e *errRetry) Error() string { return e.err.Error() }
+func (e *errRetry) Unwrap() error { return e.err }
+
+func isRetriable(err error) bool {
+	var r *errRetry
+	return errors.As(err, &r)
+}
+
+func retryWrap(err error) error {
+	if err == nil {
+		return nil
+	}
+	if isRetriable(err) {
+		return err
+	}
+	return &errRetry{err: err}
+}
+
+// Retry runs fn up to MaxAttempts times with exponential backoff between
+// attempts. Non-retriable errors abort immediately. Returns the last error.
+func Retry(ctx context.Context, policy RetryPolicy, fn func() error) error {
+	policy = policy.withDefaults()
+	var err error
+	for attempt := 1; attempt <= policy.MaxAttempts; attempt++ {
+		if err = fn(); err == nil {
+			return nil
+		}
+		if !isRetriable(err) {
+			return err
+		}
+		if attempt == policy.MaxAttempts {
+			return fmt.Errorf("after %d attempts: %w", policy.MaxAttempts, err)
+		}
+		select {
+		case <-time.After(policy.delay(attempt)):
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+	return err
+}
+
+// SyncConfig wires up a sync run.
+type SyncConfig struct {
+	OO      *OOConfig          // OnlyOffice source (optional)
+	Gmail   *GmailCredentials  // Gmail source (optional)
+	Out     string             // var/mail root; default <repo>/var/mail
+	Workers int                // concurrency; default 4
+	Limit   int                // max messages per source (0 = all)
+	Offset  int                // skip first N messages per source
+	Force   bool               // overwrite existing message.json + attachments
+	DryRun  bool               // list without writing
+	Query   string             // Gmail search query; default in:inbox
+	Policy  RetryPolicy
+}
+
+// SyncStats is returned by Run.
+type SyncStats struct {
+	Checked int
+	New     int32
+	Failed  int32
+	Skipped int32
+}
+
+// Source abstracts the two backends for the worker pool.
+type Source interface {
+	// ListIDs yields ids (string form) to fetch. cursor resumes pagination.
+	ListIDs(ctx context.Context, limit int, cursor string) (ids []string, next string, err error)
+	Get(ctx context.Context, id string) (*Message, error)
+	DownloadAttachment(ctx context.Context, msg *Message, att Attachment) ([]byte, error)
+	Folder() string
+}
+
+type ooSource struct {
+	c    *OOClient
+	page int
+}
+// gmailAPI is the Gmail client surface gmailSource needs. *GmailClient implements it.
+type gmailAPI interface {
+	ListIDs(ctx context.Context, q string, maxIDs int, pageToken string) ([]string, string, error)
+	GetMessage(ctx context.Context, id string) (*Message, error)
+	DownloadAttachment(ctx context.Context, msgID, attID string) ([]byte, error)
+}
+
+type gmailSource struct {
+	c     gmailAPI
+	cur   string
+	query string
+}
+
+func (s *ooSource) Folder() string      { return "inbox" }
+func (s *gmailSource) Folder() string   { return "gmail" }
+
+func (s *ooSource) ListIDs(ctx context.Context, limit int, cursor string) ([]string, string, error) {
+	page := s.page
+	if page == 0 {
+		page = 1
+	}
+	ids, next, err := s.c.ListIDs(ctx, limit, page)
+	s.page = next
+	strs := make([]string, len(ids))
+	for i, id := range ids {
+		strs[i] = fmt.Sprintf("%d", id)
+	}
+	return strs, "", err
+}
+
+func (s *ooSource) Get(ctx context.Context, id string) (*Message, error) {
+	var mid int
+	if _, err := fmt.Sscanf(id, "%d", &mid); err != nil {
+		return nil, fmt.Errorf("oo id %q: %w", id, err)
+	}
+	return s.c.GetMessage(ctx, mid)
+}
+
+func (s *ooSource) DownloadAttachment(ctx context.Context, msg *Message, att Attachment) ([]byte, error) {
+	return s.c.DownloadAttachment(ctx, att.FileID)
+}
+
+func (s *gmailSource) ListIDs(ctx context.Context, limit int, cursor string) ([]string, string, error) {
+	q := s.query
+	if q == "" {
+		q = "in:inbox"
+	}
+	ids, next, err := s.c.ListIDs(ctx, q, limit, cursor)
+	return ids, next, err
+}
+
+func (s *gmailSource) Get(ctx context.Context, id string) (*Message, error) {
+	return s.c.GetMessage(ctx, id)
+}
+
+func (s *gmailSource) DownloadAttachment(ctx context.Context, msg *Message, att Attachment) ([]byte, error) {
+	return s.c.DownloadAttachment(ctx, msg.ID, att.FileID)
+}
+
+// Run executes the sync across the configured sources with a worker pool.
+func Run(ctx context.Context, cfg SyncConfig) (*SyncStats, error) {
+	if cfg.Out == "" {
+		cfg.Out = "var/mail"
+	}
+	if cfg.Workers <= 0 {
+		cfg.Workers = 4
+	}
+	if err := os.MkdirAll(cfg.Out, 0o755); err != nil {
+		return nil, err
+	}
+	var sources []Source
+	if cfg.OO != nil {
+		oo, err := NewOOClient(*cfg.OO, 1) // folder inbox
+		if err != nil {
+			return nil, fmt.Errorf("onlyoffice auth: %w", err)
+		}
+		sources = append(sources, &ooSource{c: oo})
+	}
+	if cfg.Gmail != nil {
+		gm, err := NewGmailClient(*cfg.Gmail)
+		if err != nil {
+			return nil, fmt.Errorf("gmail init: %w", err)
+		}
+		sources = append(sources, &gmailSource{c: gm, query: cfg.Query})
+	}
+	if len(sources) == 0 {
+		return nil, errors.New("sync: no source configured (need OO, Gmail, or both)")
+	}
+
+	stats := &SyncStats{}
+	var jobs []struct {
+		src Source
+		id  string
+	}
+	for _, src := range sources {
+		ids, _, err := src.ListIDs(ctx, cfg.Offset+cfg.Limit, "")
+		if err != nil {
+			return nil, fmt.Errorf("list %s: %w", src.Folder(), err)
+		}
+		if cfg.Offset > 0 {
+			if cfg.Offset >= len(ids) {
+				ids = nil
+			} else {
+				ids = ids[cfg.Offset:]
+			}
+		}
+		if cfg.Limit > 0 && len(ids) > cfg.Limit {
+			ids = ids[:cfg.Limit]
+		}
+		stats.Checked += len(ids)
+		for _, id := range ids {
+			jobs = append(jobs, struct {
+				src Source
+				id  string
+			}{src: src, id: id})
+		}
+	}
+
+	var (
+		wg       sync.WaitGroup
+		mu       sync.Mutex
+		failures []string
+	)
+	jobsCh := make(chan struct {
+		src Source
+		id  string
+	})
+	for i := 0; i < cfg.Workers; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for j := range jobsCh {
+				status, err := processOne(ctx, j.src, j.id, cfg)
+				switch status {
+				case statusFailed:
+					mu.Lock()
+					failures = append(failures, j.src.Folder()+"/"+j.id+": "+err.Error())
+					mu.Unlock()
+					atomic.AddInt32(&stats.Failed, 1)
+				case statusNew:
+					atomic.AddInt32(&stats.New, 1)
+				case statusSkipped:
+					atomic.AddInt32(&stats.Skipped, 1)
+				}
+			}
+		}()
+	}
+	for _, j := range jobs {
+		select {
+		case jobsCh <- j:
+		case <-ctx.Done():
+			close(jobsCh)
+			wg.Wait()
+			return stats, ctx.Err()
+		}
+	}
+	close(jobsCh)
+	wg.Wait()
+
+	if len(failures) > 0 {
+		fmt.Fprintf(os.Stderr, "sync: %d failures:\n  %s\n", len(failures), strings.Join(failures, "\n  "))
+	}
+	return stats, nil
+}
+
+type status int
+
+const (
+	statusNew status = iota
+	statusSkipped
+	statusFailed
+)
+
+func processOne(ctx context.Context, src Source, id string, cfg SyncConfig) (status, error) {
+	if cfg.DryRun {
+		return statusNew, nil
+	}
+	dir := filepath.Join(cfg.Out, src.Folder(), id)
+	jsonPath := filepath.Join(dir, "message.json")
+	if !cfg.Force {
+		if _, err := os.Stat(jsonPath); err == nil {
+			return statusSkipped, nil
+		}
+	}
+	var msg *Message
+	err := Retry(ctx, cfg.Policy, func() error {
+		m, err := src.Get(ctx, id)
+		if err != nil {
+			return retryWrap(err)
+		}
+		m.Folder = src.Folder() // directory layout is authoritative
+		if err := writeMessage(jsonPath, m); err != nil {
+			return err
+		}
+		msg = m
+		return nil
+	})
+	if err != nil {
+		return statusFailed, err
+	}
+	for _, att := range msg.Attachments {
+		attDir := filepath.Join(dir, "attachments")
+		if err := os.MkdirAll(attDir, 0o755); err != nil {
+			return statusFailed, err
+		}
+		attPath := filepath.Join(attDir, sanitize(att.StoredName))
+		if _, err := os.Stat(attPath); err == nil && !cfg.Force {
+			continue
+		}
+		var data []byte
+		err := Retry(ctx, cfg.Policy, func() error {
+			b, err := src.DownloadAttachment(ctx, msg, att)
+			if err != nil {
+				return retryWrap(err)
+			}
+			data = b
+			return os.WriteFile(attPath, b, 0o644)
+		})
+		if err != nil {
+			return statusFailed, fmt.Errorf("attachment %s: %w", att.FileName, err)
+		}
+		// ICS attachments get structured markdown immediately (same name the
+		// Python converter would use: <display stem>.md).
+		if isICS(att.FileName) {
+			stem := att.FileName
+			if i := strings.LastIndex(stem, "."); i >= 0 {
+				stem = stem[:i]
+			}
+			mdPath := filepath.Join(attDir, sanitize(stem)+".md")
+			if err := os.WriteFile(mdPath, []byte(ICSToMarkdown(data)), 0o644); err != nil {
+				return statusFailed, err
+			}
+		}
+	}
+	return statusNew, nil
+}
+
+func writeMessage(path string, m *Message) error {
+	b, err := json.MarshalIndent(m, "", "  ")
+	if err != nil {
+		return err
+	}
+	if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
+		return err
+	}
+	return os.WriteFile(path, b, 0o644)
+}
+
+func sanitize(name string) string {
+	r := strings.NewReplacer("/", "_", "\\", "_", ":", "_", "*", "_", "?", "_", "\"", "_",
+		"<", "_", ">", "_", "|", "_", " ", "_")
+	return r.Replace(name)
+}
+
+func isICS(name string) bool {
+	n := strings.ToLower(name)
+	return strings.HasSuffix(n, ".ics") || strings.HasSuffix(n, ".ical")
+}
@@ -0,0 +1,305 @@
+package sync
+
+import (
+	"context"
+	"encoding/base64"
+	"errors"
+	"testing"
+	"time"
+	"unicode/utf8"
+)
+
+func TestRetrySucceedsOnSecondTry(t *testing.T) {
+	attempts := 0
+	err := Retry(context.Background(), RetryPolicy{BaseDelay: time.Millisecond, MaxDelay: 5 * time.Millisecond}, func() error {
+		attempts++
+		if attempts == 1 {
+			return retryWrap(errors.New("boom"))
+		}
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+	if attempts != 2 {
+		t.Fatalf("expected 2 attempts, got %d", attempts)
+	}
+}
+
+func TestRetryExhaustsAttempts(t *testing.T) {
+	attempts := 0
+	err := Retry(context.Background(), RetryPolicy{MaxAttempts: 3, BaseDelay: time.Millisecond, MaxDelay: 5 * time.Millisecond}, func() error {
+		attempts++
+		return retryWrap(errors.New("nope"))
+	})
+	if err == nil {
+		t.Fatal("expected error after exhaustion")
+	}
+	if attempts != 3 {
+		t.Fatalf("expected 3 attempts, got %d", attempts)
+	}
+}
+
+func TestRetryNonRetriableAbortsImmediately(t *testing.T) {
+	attempts := 0
+	err := Retry(context.Background(), RetryPolicy{MaxAttempts: 5, BaseDelay: time.Millisecond}, func() error {
+		attempts++
+		return errors.New("permanent")
+	})
+	if err == nil {
+		t.Fatal("expected error")
+	}
+	if attempts != 1 {
+		t.Fatalf("expected 1 attempt for non-retriable, got %d", attempts)
+	}
+}
+
+func TestRetryRespectsContext(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	err := Retry(ctx, RetryPolicy{MaxAttempts: 5, BaseDelay: time.Millisecond}, func() error {
+		return retryWrap(errors.New("x"))
+	})
+	if !errors.Is(err, context.Canceled) {
+		t.Fatalf("expected context.Canceled, got %v", err)
+	}
+}
+
+func TestDelayGrows(t *testing.T) {
+	p := RetryPolicy{BaseDelay: time.Second, MaxDelay: 30 * time.Second, Jitter: 0}
+	d1 := p.delay(1) // attempt 1 => 0
+	d2 := p.delay(2)
+	d3 := p.delay(3)
+	if d1 != 0 {
+		t.Fatalf("attempt 1 delay should be 0, got %v", d1)
+	}
+	if d2 != time.Second {
+		t.Fatalf("attempt 2 delay should be 1s, got %v", d2)
+	}
+	if d3 != 2*time.Second {
+		t.Fatalf("attempt 3 delay should be 2s, got %v", d3)
+	}
+}
+
+func TestSanitize(t *testing.T) {
+	cases := map[string]string{
+		"a/b\\c:d*e": "a_b_c_d_e",
+		"normal.txt": "normal.txt",
+		"../evil":    ".._evil",
+		"a b c.pdf":  "a_b_c.pdf",
+	}
+	for in, want := range cases {
+		if got := sanitize(in); got != want {
+			t.Errorf("sanitize(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+func TestIsICS(t *testing.T) {
+	if !isICS("reply.ics") || !isICS("x.ICAL") {
+		t.Fatal("ics extensions not detected")
+	}
+	if isICS("invoice.pdf") {
+		t.Fatal("pdf misdetected as ics")
+	}
+}
+
+const fixtureReplyICS = `BEGIN:VCALENDAR
+METHOD:REPLY
+PRODID:Microsoft Exchange Server 2010
+VERSION:2.0
+BEGIN:VTIMEZONE
+TZID:W. Europe Standard Time
+BEGIN:STANDARD
+DTSTART:16010101T030000
+TZOFFSETFROM:+0200
+TZOFFSETTO:+0100
+RRULE:FREQ=YEARLY;INTERVAL=1;BYDAY=-1SU;BYMONTH=10
+END:STANDARD
+BEGIN:DAYLIGHT
+DTSTART:16010101T020000
+TZOFFSETFROM:+0100
+TZOFFSETTO:+0200
+RRULE:FREQ=YEARLY;INTERVAL=1;BYDAY=-1SU;BYMONTH=3
+END:DAYLIGHT
+END:VTIMEZONE
+BEGIN:VEVENT
+ATTENDEE;PARTSTAT=ACCEPTED;CN="Baker, Ben":mailto:bbaker1@teksystems.com
+UID:bvlnr1i35ug30kn6rvu9dop00g@google.com
+SUMMARY;LANGUAGE=en-US:Accepted: Appointment (Ben  Baker)
+DTSTART;TZID=W. Europe Standard Time:20260812T120000
+DTEND;TZID=W. Europe Standard Time:20260812T123000
+CLASS:PUBLIC
+STATUS:CONFIRMED
+LOCATION;LANGUAGE=en-US:https://meet.google.com/sxh-ubud-jrd
+END:VEVENT
+END:VCALENDAR`
+
+func TestICSToMarkdown(t *testing.T) {
+	out := ICSToMarkdown([]byte(fixtureReplyICS))
+	for _, want := range []string{
+		"Accepted: Appointment",
+		"When:",
+		"Where:",
+		"meet.google.com",
+		"Attendee:",
+		"Baker, Ben",
+		"Accepted",
+		"Calendar method: REPLY",
+	} {
+		if !contains(out, want) {
+			t.Errorf("output missing %q:\n%s", want, out)
+		}
+	}
+	if contains(out, "BEGIN:VCALENDAR") {
+		t.Errorf("raw ICS leaked into markdown:\n%s", out)
+	}
+}
+
+func TestICSToMarkdownFallback(t *testing.T) {
+	out := ICSToMarkdown([]byte("not a calendar"))
+	if !contains(out, "not a calendar") {
+		t.Fatalf("expected raw fallback, got %q", out)
+	}
+}
+
+func TestICSToMarkdownNormalizesLatin1(t *testing.T) {
+	// Real-world ICS from a rental portal: summary in UTF-8, location Latin-1
+	// ("N\xfcrnberg"). The markdown output must be valid UTF-8 everywhere.
+	raw := "BEGIN:VCALENDAR\r\nVERSION:2.0\r\nBEGIN:VEVENT\r\n" +
+		"SUMMARY:Mietwagen-Buchung: N\xc3\xbcrnberg\r\n" +
+		"LOCATION:N\xfcrnberg\r\nDTSTART:20200101T090000Z\r\nDTEND:20200101T180000Z\r\n" +
+		"END:VEVENT\r\nEND:VCALENDAR\r\n"
+	out := ICSToMarkdown([]byte(raw))
+	if !utf8.ValidString(out) {
+		t.Fatalf("output is not valid UTF-8:\n%q", out)
+	}
+	if !contains(out, "Nürnberg") {
+		t.Errorf("expected Nürnberg in output:\n%s", out)
+	}
+	if contains(out, "N\xfcrnberg") {
+		t.Errorf("Latin-1 bytes leaked into output:\n%q", out)
+	}
+}
+
+func TestICSToMarkdownAllDay(t *testing.T) {
+	ics := `BEGIN:VCALENDAR
+VERSION:2.0
+BEGIN:VEVENT
+UID:y@google.com
+SUMMARY:All day thing
+DTSTART;VALUE=DATE:20260815
+DTEND;VALUE=DATE:20260816
+END:VEVENT
+END:VCALENDAR`
+	out := ICSToMarkdown([]byte(ics))
+	if !contains(out, "All day thing") || !contains(out, "2026-08-15") {
+		t.Errorf("all-day event not parsed:\n%s", out)
+	}
+}
+
+func TestCollectParts(t *testing.T) {
+	p := gmailPart{
+		MimeType: "multipart/mixed",
+		Parts: []gmailPart{
+			{MimeType: "multipart/alternative", Parts: []gmailPart{
+				{MimeType: "text/plain", Body: gmailBody{Data: b64("plain text")}},
+				{MimeType: "text/html", Body: gmailBody{Data: b64("<p>html</p>")}},
+			}},
+			{PartID: "2", MimeType: "application/pdf", Filename: "invoice.pdf", Body: gmailBody{Size: 100}},
+		},
+	}
+	text, html, atts := collectParts(p, "root", "abc123", 0)
+	if text != "plain text" {
+		t.Errorf("text = %q", text)
+	}
+	if html != "<p>html</p>" {
+		t.Errorf("html = %q", html)
+	}
+	if len(atts) != 1 || atts[0].FileName != "invoice.pdf" || atts[0].FileID != "abc123:2" {
+		t.Errorf("atts = %+v", atts)
+	}
+}
+
+type fakeGmailAPI struct {
+	lastQ     string
+	lastLimit int
+	ids       []string
+}
+
+func (f *fakeGmailAPI) ListIDs(_ context.Context, q string, maxIDs int, _ string) ([]string, string, error) {
+	f.lastQ = q
+	f.lastLimit = maxIDs
+	return f.ids, "", nil
+}
+func (f *fakeGmailAPI) GetMessage(context.Context, string) (*Message, error) {
+	return nil, errors.New("unused")
+}
+func (f *fakeGmailAPI) DownloadAttachment(context.Context, string, string) ([]byte, error) {
+	return nil, errors.New("unused")
+}
+
+func TestGmailSourcePassesQueryToListIDs(t *testing.T) {
+	fake := &fakeGmailAPI{ids: []string{"m1"}}
+	src := &gmailSource{c: fake, query: "from:alice@example.com"}
+	ids, _, err := src.ListIDs(context.Background(), 10, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if fake.lastQ != "from:alice@example.com" {
+		t.Fatalf("ListIDs q=%q, want from:alice@example.com", fake.lastQ)
+	}
+	if fake.lastLimit != 10 {
+		t.Fatalf("ListIDs limit=%d, want 10", fake.lastLimit)
+	}
+	if len(ids) != 1 || ids[0] != "m1" {
+		t.Fatalf("ids=%v", ids)
+	}
+}
+
+func TestGmailSourceEmptyQueryDefaultsToInbox(t *testing.T) {
+	fake := &fakeGmailAPI{}
+	src := &gmailSource{c: fake, query: ""}
+	if _, _, err := src.ListIDs(context.Background(), 5, ""); err != nil {
+		t.Fatal(err)
+	}
+	if fake.lastQ != "in:inbox" {
+		t.Fatalf("empty query q=%q, want in:inbox", fake.lastQ)
+	}
+}
+
+func TestParseCLIGmailQuery(t *testing.T) {
+	cli, code, err := ParseCLI([]string{
+		"--source", "gmail",
+		"--query", "from:letrado@example.com",
+		"--out", t.TempDir(),
+		"--dry-run",
+	})
+	if err != nil || code != 0 {
+		t.Fatalf("ParseCLI: code=%d err=%v", code, err)
+	}
+	if cli.Sync.Query != "from:letrado@example.com" {
+		t.Fatalf("query=%q", cli.Sync.Query)
+	}
+	if cli.Sync.Gmail == nil {
+		t.Fatal("gmail source not configured")
+	}
+}
+
+func b64(s string) string {
+	return base64.URLEncoding.EncodeToString([]byte(s))
+}
+
+func contains(s, sub string) bool {
+	return len(s) >= len(sub) && (s == sub || len(sub) == 0 ||
+		indexOf(s, sub) >= 0)
+}
+
+func indexOf(s, sub string) int {
+	for i := 0; i+len(sub) <= len(s); i++ {
+		if s[i:i+len(sub)] == sub {
+			return i
+		}
+	}
+	return -1
+}
@@ -0,0 +1,46 @@
+// Package sync downloads OnlyOffice and Gmail messages to var/mail/ as raw
+// JSON + attachment files, then hands off to bin/mail/import --from-raw for
+// markdown conversion.
+//
+// On-disk schema (per message):
+//
+//	var/mail/<folder>/<id>/message.json   # Message (this package)
+//	var/mail/<folder>/<id>/attachments/    # raw attachment bytes (storedName)
+//
+// The Message JSON is the contract shared with the Python converter. Fields
+// deliberately mirror what bin/mail/import already reads from the OnlyOffice
+// API, so conversion is source-agnostic.
+package sync
+
+import (
+	"time"
+)
+
+// Attachment describes one attachment of a Message. FileID/FileName/StoredName
+// mirror OnlyOffice; Gmail fills them from its own ids. StoredName is always
+// unique (hash/attachment id) so raw files never collide.
+type Attachment struct {
+	FileID      string `json:"fileId,omitempty"`
+	FileName    string `json:"fileName"`
+	StoredName  string `json:"storedName"`
+	Size        int64  `json:"size,omitempty"`
+	ContentType string `json:"contentType,omitempty"`
+}
+
+// Message is the normalized record written to var/mail/<folder>/<id>/message.json.
+type Message struct {
+	Source         string       `json:"source"` // "onlyoffice" | "gmail"
+	ID             string       `json:"id"`
+	Folder         string       `json:"folder"`
+	Subject        string       `json:"subject,omitempty"`
+	From           string       `json:"from,omitempty"`
+	To             string       `json:"to,omitempty"`
+	CC             string       `json:"cc,omitempty"`
+	BCC            string       `json:"bcc,omitempty"`
+	ReceivedAt     time.Time    `json:"receivedAt,omitempty"`
+	HTMLBody       string       `json:"htmlBody,omitempty"`
+	TextBody       string       `json:"textBody,omitempty"`
+	HasAttachments bool         `json:"hasAttachments,omitempty"`
+	Attachments    []Attachment `json:"attachments,omitempty"`
+	MimeMessageID  string       `json:"mimeMessageId,omitempty"`
+}
@@ -0,0 +1,3 @@
+// Commands in this directory are shebang mains (import.go), tagged so
+// `go build ./bin/markdown` does not see two mains.
+package main
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=markdown_import "$0" "$@"; exit
+//go:build markdown_import
+//
+// bin/markdown/import.go - split markdown into leafs (mistune).
+//
+//	./bin/markdown/import.go [dir]
+//	./bin/markdown/import.go --files a.md,b.md --json
+//
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/md/import", os.Args[1:]))
+}
@@ -1,9 +1,10 @@
 #!/usr/bin/env python3
-import lib
+from __future__ import annotations
+
 import sys
 from pathlib import Path

-sys.path.insert(0, str(Path(__file__).resolve().parent))
+sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "tools"))

 from mdleaves import leaves_to_json, read_markdown, to_all, walk_markdown  # noqa: E402
 from yamlout import to_yaml  # noqa: E402
@@ -0,0 +1,2 @@
+// Commands in this directory are shebang mains (query.go).
+package main
@@ -0,0 +1,20 @@
+//usr/bin/env go run -tags=postgres_query "$0" "$@"; exit
+//go:build postgres_query
+//
+// bin/postgres/query.go - read-only Postgres as YAML.
+//
+//	./bin/postgres/query.go --profile onlyoffice -c 'SELECT 1'
+//
+// Profiles: $HOME/.config/brain/db-profiles.yml (credentials stay out of git).
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+package main
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/cmdbin"
+)
+
+func main() {
+	os.Exit(cmdbin.ExecFile("bin/db/psql-yq", os.Args[1:]))
+}
@@ -0,0 +1,22 @@
+//usr/bin/env go run -tags=brain_serve "$0" "$@"; exit
+//go:build brain_serve
+//
+// bin/serve.go — deprecated; use bin/brain/serve.go.
+package main
+
+import (
+	"fmt"
+	"os"
+
+	"github.com/eSlider/2dph/internal/httpapi"
+)
+
+func main() {
+	fmt.Fprintln(os.Stderr, "bin/serve.go is deprecated; use bin/brain/serve.go")
+	if os.Getenv("KB_ROOT") == "" {
+		if wd, err := os.Getwd(); err == nil {
+			os.Setenv("KB_ROOT", wd)
+		}
+	}
+	httpapi.Run(nil)
+}
@@ -0,0 +1,29 @@
+"""crmfacts - pure helpers for bin/facts/crm (association proofing).
+
+Shared with tools/ unit tests so the corpus-org parser is covered in CI.
+"""
+
+import re
+
+
+def corpus_orgs(raw: str) -> dict[str, dict]:
+    """Parse the orgs block of the CV knowledge-mesh YAML into id -> fields.
+
+    Fields kept: label, kind, period, website. Stops at the first sibling
+    top-level key (clients, timeline, ...).
+    """
+    m = re.search(r"^orgs:\n(.*?)\n^(?:clients|timeline|tech_weights|nodes|edges):", raw, re.S | re.M)
+    if not m:
+        return {}
+    orgs: dict[str, dict] = {}
+    cur = None
+    for line in m.group(1).splitlines():
+        lm = re.match(r"^\s*- id:\s*(\S+)", line)
+        if lm:
+            cur = lm.group(1)
+            orgs[cur] = {}
+            continue
+        fm = re.match(r"^\s+(\w+):\s*(.*)$", line)
+        if fm and cur and fm.group(1) in ("label", "kind", "period", "website"):
+            orgs[cur][fm.group(1)] = fm.group(2).strip()
+    return orgs
@@ -0,0 +1,117 @@
+"""gitimport - parse `git log` output and turn commits into brain leafs.
+
+Pure, testable functions. Field grammar (see bin/git/import):
+
+  git log --no-merges --name-only \
+      --format='%x1e%H%x1f%an%x1f%ae%x1f%aI%x1f%s'
+
+  0x1e = record separator, 0x1f = field separator.
+  Files: newline-separated lines following each record's subject.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+REC_SEP = "\x1e"
+FIELD_SEP = "\x1f"
+
+
+@dataclass
+class Commit:
+    sha: str
+    author: str
+    email: str
+    date: str
+    subject: str
+    files: list[str] = field(default_factory=list)
+
+    def leaf_text(self, repo: str) -> str:
+        head = f"commit {self.sha[:12]} in {repo} — {self.subject}"
+        body = [head, f"Author: {self.author} <{self.email}>", f"Date: {self.date}"]
+        if self.files:
+            body.append("Changing: " + ", ".join(self.files))
+        return "\n".join(body)
+
+
+def parse_log(text: str) -> list[Commit]:
+    """Parse `git log` output into Commit records.
+
+    Records are separated by 0x1e. A record is fields joined by 0x1f,
+    followed by optional newline-separated file paths inside the next
+    segment (git emits blank line + files after each record).
+    """
+    commits: list[Commit] = []
+    # field records and file lists alternate; simpler: split on REC_SEP,
+    # each chunk = header line, possibly followed by newline + files.
+    for chunk in text.split(REC_SEP):
+        chunk = chunk.strip("\n")
+        if not chunk:
+            continue
+        lines = chunk.split("\n", 1)
+        header = lines[0].split(FIELD_SEP)
+        if len(header) < 5:
+            continue
+        sha, author, email, date, subject = header[:5]
+        files = [ln.strip() for ln in lines[1].splitlines() if ln.strip()] if len(lines) > 1 else []
+        commits.append(Commit(sha=sha, author=author, email=email,
+                              date=date, subject=subject, files=files))
+    return commits
+
+
+def commits_to_leafs(commits: list[Commit], repo: str) -> list[dict]:
+    """Map commits to the leaf shape bin/kb/index expects (source/repo/...)."""
+    out: list[dict] = []
+    for c in commits:
+        out.append({
+            "source": f"{repo}@{c.sha}",
+            "repo": repo,
+            "heading": f"commit {c.sha[:12]} — {c.subject}",
+            "text": c.leaf_text(repo),
+            "type": "commit",
+            "status": "current",
+            "related": ",".join(c.files),
+        })
+    return out
+
+
+GIT_SCHEMA = (
+    "CREATE NODE TABLE IF NOT EXISTS Commit (id STRING, repo STRING, subject STRING, "
+    "author STRING, email STRING, date STRING, PRIMARY KEY(id))",
+    "CREATE NODE TABLE IF NOT EXISTS Person (id STRING, name STRING, email STRING, PRIMARY KEY(id))",
+    "CREATE REL TABLE IF NOT EXISTS HAS_VERSION (FROM File TO Commit)",
+    "CREATE REL TABLE IF NOT EXISTS AUTHORED (FROM Commit TO Person)",
+)
+
+
+def ensure_git_schema(conn) -> None:
+    for stmt in GIT_SCHEMA:
+        conn.execute(stmt)
+
+
+def index_commits(conn, commits: list[Commit], repo: str) -> int:
+    """Write Commit/File/Person nodes + edges, one per commit (idempotent by sha)."""
+    ensure_git_schema(conn)
+    for c in commits:
+        conn.execute(
+            "MERGE (c:Commit {id:$sha}) SET c.repo=$repo, c.subject=$subject, "
+            "c.author=$author, c.email=$email, c.date=$date",
+            parameters={"sha": c.sha, "repo": repo, "subject": c.subject,
+                        "author": c.author, "email": c.email, "date": c.date},
+        )
+        conn.execute(
+            "MERGE (p:Person {id:$email}) SET p.name=$name, p.email=$email",
+            parameters={"email": c.email, "name": c.author},
+        )
+        conn.execute("MATCH (c:Commit {id:$sha}), (p:Person {id:$email}) "
+                     "MERGE (c)-[:AUTHORED]->(p)",
+                     parameters={"sha": c.sha, "email": c.email})
+        for path in c.files:
+            conn.execute(
+                "MERGE (f:File {id:$fid}) SET f.path=$path, f.repo=$repo",
+                parameters={"fid": f"{repo}:{path}", "path": path, "repo": repo},
+            )
+            conn.execute("MATCH (f:File {id:$fid}), (c:Commit {id:$sha}) "
+                         "MERGE (f)-[:HAS_VERSION]->(c)",
+                         parameters={"fid": f"{repo}:{path}", "sha": c.sha})
+    return len(commits)
@@ -22,7 +22,17 @@ ROOT_FACTS = "facts"
 ROOT_INFO = "info"
 CONF_CONFIRMED = "confirmed"

-VAR = Path(__file__).resolve().parents[1] / "var"
+def _repo_root() -> Path:
+    p = Path(__file__).resolve().parent
+    while True:
+        if (p / "var").is_dir() or (p / ".git").is_dir() or (p / "pyproject.toml").is_file():
+            return p
+        if p.parent == p:
+            return Path(__file__).resolve().parents[2]
+        p = p.parent
+
+
+VAR = _repo_root() / "var"
 DB_PATH = VAR / "kb.lbug"


@@ -68,6 +78,19 @@ def init_schema(conn: ladybug.Connection) -> None:
    conn.execute(
        "CREATE REL TABLE IF NOT EXISTS RUNS_ON (FROM Leaf TO Host)"
    )
+    conn.execute(
+        "CREATE NODE TABLE IF NOT EXISTS Commit (id STRING, repo STRING, subject STRING, "
+        "author STRING, email STRING, date STRING, PRIMARY KEY(id))"
+    )
+    conn.execute(
+        "CREATE NODE TABLE IF NOT EXISTS Person (id STRING, name STRING, email STRING, PRIMARY KEY(id))"
+    )
+    conn.execute(
+        "CREATE REL TABLE IF NOT EXISTS HAS_VERSION (FROM File TO Commit)"
+    )
+    conn.execute(
+        "CREATE REL TABLE IF NOT EXISTS AUTHORED (FROM Commit TO Person)"
+    )


 def leaf_id(text: str, source: str) -> str:
@@ -95,18 +118,77 @@ def upsert_leaf(conn: ladybug.Connection, *, text: str, root: str, confidence: s
    return lid


+def leaf_index_names(conn: ladybug.Connection) -> set[str]:
+    """Return index names on the Leaf table (e.g. {'id', 'Leaf_vec', '_PK'})."""
+    rows = conn.execute("CALL SHOW_INDEXES() RETURN *").get_all()
+    return {row[1] for row in rows if row[0] == "Leaf"}
+
+
 def create_fts_and_vector(conn: ladybug.Connection, force: bool = False) -> None:
-    if force:
-        conn.execute("DROP INDEX IF EXISTS Leaf.Leaf_fts")
-        conn.execute("DROP INDEX IF EXISTS Leaf.Leaf_vec")
-    try:
-        conn.execute("CALL CREATE_FTS_INDEX('Leaf', 'id', ['text'])")
-    except Exception:
-        pass
-    try:
-        conn.execute("CALL CREATE_VECTOR_INDEX('Leaf', 'Leaf_vec', 'embedding', metric := 'cosine')")
-    except Exception:
-        pass
+    """Create FTS (BM25) + HNSW vector indexes if missing.
+
+    Never DROP INDEX for FTS/VECTOR. Ladybug 0.19 leaves ghost catalog
+    entries after DROP (`_0_Leaf_vec_UPPER`, `0_id_docs`), so a later
+    CREATE fails with "already exists in catalog" while SHOW_INDEXES
+    still omits the index. Swallowing that error made HNSW look "OK"
+    until the first QUERY_VECTOR_INDEX.
+
+    `force=True` is accepted for API compatibility but does **not** drop.
+    Fresh indexes require deleting `var/kb.lbug` and rebuilding
+    (`bin/brain/index.go --rebuild`).
+    """
+    del force  # API compat; DROP is unsafe — see docstring
+    names = leaf_index_names(conn)
+    if "id" not in names:
+        try:
+            conn.execute("CALL CREATE_FTS_INDEX('Leaf', 'id', ['text'])")
+        except Exception as e:
+            raise RuntimeError(
+                "CREATE_FTS_INDEX failed (often ghost catalog after DROP INDEX). "
+                "Delete var/kb.lbug and run bin/brain/index.go --rebuild. "
+                f"Cause: {e}"
+            ) from e
+    if "Leaf_vec" not in names:
+        try:
+            conn.execute(
+                "CALL CREATE_VECTOR_INDEX('Leaf', 'Leaf_vec', 'embedding', "
+                "metric := 'cosine')"
+            )
+        except Exception as e:
+            raise RuntimeError(
+                "CREATE_VECTOR_INDEX failed (often ghost catalog after DROP INDEX "
+                "Leaf.Leaf_vec → `_0_Leaf_vec_UPPER already exists in catalog`). "
+                "Delete var/kb.lbug and run bin/brain/index.go --rebuild. "
+                f"Cause: {e}"
+            ) from e
+    names = leaf_index_names(conn)
+    missing = {"id", "Leaf_vec"} - names
+    if missing:
+        raise RuntimeError(
+            f"Leaf indexes incomplete after create: missing {sorted(missing)}; "
+            f"have {sorted(names)}. Delete var/kb.lbug and --rebuild."
+        )
+
+
+def ensure_indexes(conn: ladybug.Connection) -> None:
+    """Idempotent: create FTS + HNSW only when missing. Safe after upserts."""
+    create_fts_and_vector(conn, force=False)
+
+
+def drop_indexes(conn: ladybug.Connection) -> None:
+    """No-op. Kept for callers; DROP INDEX is fatal on Ladybug 0.19.
+
+    Historical note claimed "drop before bulk MERGE". Measured on 0.19:
+    - DROP FTS/VECTOR leaves ghost catalog → CREATE fails permanently until
+      `var/kb.lbug` is deleted.
+    - MERGE/upsert while **FTS** exists can corrupt FTS
+      ("document for node offset N is missing during delete").
+    - Upsert while **HNSW** exists stays queryable.
+
+    Bulk rebuilders must delete `var/kb.lbug`, write all leafs (info+facts)
+    with no indexes, then `ensure_indexes()` once.
+    """
+    return


 def query_fts(conn: ladybug.Connection, text: str, limit: int = 10) -> list[dict]:
@@ -155,6 +237,6 @@ def stats(conn: ladybug.Connection) -> dict:

 def open_readonly() -> tuple[ladybug.Database, ladybug.Connection]:
    if not DB_PATH.exists():
-        raise FileNotFoundError(f"{DB_PATH} missing - run bin/kb/index first")
+        raise FileNotFoundError(f"{DB_PATH} missing - run bin/brain/index.go --rebuild first")
    db, conn = connect(read_only=True)
    return db, conn
@@ -0,0 +1,148 @@
+"""mailconv - pure helpers for bin/mail/import (mail -> markdown + attachments).
+
+Shared with unit tests in bin/tools/test_mailconv.py. No network, no OnlyOffice
+dependencies here: everything is `str -> str` or `Path -> str` so the tests run
+offline against fixtures.
+"""
+from __future__ import annotations
+
+import html
+import re
+import zipfile
+from pathlib import Path
+
+# Body part / attachment file suffixes we know how to turn into markdown text.
+TEXT_SUFFIXES = {".md", ".markdown", ".txt", ".csv", ".json", ".xml", ".yaml", ".yml", ".log", ".tsv",
+                 ".ics", ".ical", ".vcf", ".eml"}
+OFFICE_SUFFIXES = {".docx", ".pptx", ".xlsx", ".html", ".htm", ".epub", ".eml", ".msg"}
+PDF_SUFFIXES = {".pdf"}
+IMAGE_SUFFIXES = {".png", ".jpg", ".jpeg", ".gif", ".bmp", ".tiff", ".tif", ".webp"}
+ARCHIVE_SUFFIXES = {".zip"}
+# Legacy binary Office (doc/xls/ppt) — markitdown/docling skip them; we try
+# pandoc first, else leave a stub.
+LEGACY_OFFICE_SUFFIXES = {".doc", ".xls", ".ppt"}
+
+CONVERTIBLE_SUFFIXES = (
+    TEXT_SUFFIXES | OFFICE_SUFFIXES | PDF_SUFFIXES | IMAGE_SUFFIXES | ARCHIVE_SUFFIXES | LEGACY_OFFICE_SUFFIXES
+)
+
+
+def clean_email_address(raw: str) -> str:
+    """Extract the bare email from '"Name" <a@b.c>' and strip control chars."""
+    m = re.search(r"<([^<>@\s]+@[^<>@\s]+)>", raw)
+    return (m.group(1) if m else raw).strip()
+
+
+def subject_to_filename(subject: str, max_len: int = 80) -> str:
+    """Turn a mail subject into a filesystem-safe slug (keep first token readable)."""
+    s = re.sub(r"[^\w\-. ]+", "", subject).strip()
+    s = re.sub(r"\s+", "_", s)
+    s = s.strip("._")
+    if not s:
+        s = "untitled"
+    return s[:max_len] or "untitled"
+
+
+def strip_html(html_text: str) -> str:
+    """Naive HTML -> plain text fallback (used only if markitdown is missing)."""
+    import re as _re
+    text = _re.sub(r"(?is)<(script|style)[^>]*>.*?</\1>", "", html_text)
+    text = _re.sub(r"(?s)<br\s*/?>", "\n", text)
+    text = _re.sub(r"(?s)</p>", "\n\n", text)
+    text = _re.sub(r"(?s)<[^>]+>", "", text)
+    return html.unescape(text).strip()
+
+
+def _unwrap_tables(html_text: str) -> str:
+    """Unwrap mail HTML tables into pipe-joined text lines.
+
+    Outlook/Stripe-style emails wrap content in nested spacer/frame tables that
+    markitdown renders as hundreds of `--- |` cells and duplicated blocks.
+    Every <table> becomes plain "cell1 | cell2" lines (key-value pairs survive),
+    so only headings/paragraphs/links reach markitdown and no table noise is left.
+    """
+    try:
+        from bs4 import BeautifulSoup
+    except Exception:
+        return html_text
+    soup = BeautifulSoup(html_text, "html.parser")
+    for table in reversed(soup.find_all("table")):
+        lines: list[str] = []
+        for row in table.find_all("tr"):
+            cells = [c.get_text(" ", strip=True) for c in row.find_all(["td", "th"])]
+            line = " | ".join(x for x in cells if x)
+            if line:
+                lines.append(line)
+        if lines:
+            table.replace_with(BeautifulSoup("\n".join(lines), "html.parser"))
+        else:
+            table.decompose()
+    return str(soup)
+
+
+def html_to_markdown(html_text: str) -> str:
+    """Convert a mail HTML body to markdown using markitdown when available."""
+    html_text = _unwrap_tables(html_text)
+    try:
+        from markitdown import MarkItDown
+        import io
+        md = MarkItDown()
+        result = md.convert_stream(io.BytesIO(html_text.encode("utf-8", errors="replace")),
+                                   file_extension=".html")
+        text = result.text_content.strip()
+        if text:
+            return normalize_markdown(text)
+    except Exception:
+        pass
+    return normalize_markdown(strip_html(html_text))
+
+
+def normalize_markdown(text: str) -> str:
+    """Collapse the pdfminer/markitdown NUL artifacts and stray control chars."""
+    # NUL bytes that pdfminer inserts between digits/letters.
+    text = text.replace("\x00", "")
+    # Email spacer noise: zero-width chars, soft hyphens, figure spaces,
+    # combining grapheme joiner, BOM.
+    for ch in ("\ufeff", "\u200b", "\u034f", "\u00ad", "\u2007", "\u2008", "\u200a", "\u2002"):
+        text = text.replace(ch, "")
+    text = re.sub(r"[ \t]{2,}", " ", text)
+    # Trim trailing whitespace per line so space-only spacer rows collapse.
+    text = "\n".join(l.rstrip() for l in text.split("\n"))
+    # Collapse 3+ blank lines to two.
+    text = re.sub(r"\n{3,}", "\n\n", text)
+    # Remove weird trailing control chars.
+    text = "".join(ch for ch in text if ch >= " " or ch in "\n\t")
+    return text.strip()
+
+
+def split_zip_members(zip_path: Path) -> list[str]:
+    """Return safe member names of a zip archive (skips dir entries)."""
+    try:
+        with zipfile.ZipFile(zip_path) as zf:
+            return [m for m in zf.namelist() if not m.endswith("/")]
+    except zipfile.BadZipFile:
+        return []
+
+
+def zip_extract_safe(zip_path: Path, dest: Path) -> list[Path]:
+    """Extract a zip into dest guarding against path traversal; returns files."""
+    out: list[Path] = []
+    try:
+        with zipfile.ZipFile(zip_path) as zf:
+            for member in zf.infolist():
+                if member.is_dir():
+                    continue
+                target = (dest / member.filename).resolve()
+                if not target.is_relative_to(dest.resolve()):
+                    continue
+                target.parent.mkdir(parents=True, exist_ok=True)
+                with zf.open(member) as src, open(target, "wb") as dst:
+                    dst.write(src.read())
+                out.append(target)
+    except zipfile.BadZipFile:
+        return []
+    return out
+
+
+def is_convertible(suffix: str) -> bool:
+    return suffix.lower() in CONVERTIBLE_SUFFIXES
@@ -0,0 +1,37 @@
+"""Mail markdown under var/mail → info leafs. Conversion stays off the brain DB."""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from mdleaves import read_markdown, to_all
+
+
+def msg_date(md: Path) -> str:
+    j = md.parent / "message.json"
+    try:
+        d = json.loads(j.read_text(encoding="utf-8"))
+        return (d.get("receivedDate") or d.get("receivedAt") or "")[:10]
+    except (OSError, json.JSONDecodeError, TypeError):
+        return ""
+
+
+def from_mail_root(root: Path, limit: int = 0, since: str = "", repo: str = "ooMail") -> list[dict]:
+    if not root.is_dir():
+        return []
+    mds = sorted(root.rglob("message.md"))
+    if since:
+        mds = [m for m in mds if msg_date(m) >= since]
+    if limit:
+        mds = mds[:limit]
+    leafs: list[dict] = []
+    for md in mds:
+        files = [md] + sorted((md.parent / "attachments").glob("*.md"))
+        for f in files:
+            if not f.exists():
+                continue
+            for lf in to_all(read_markdown(f), f, repo=repo):
+                lf["source"] = f"ooMail:{md.parent.name}:{f.name}"
+                lf["how"] = "mail/import"
+                leafs.append(lf)
+    return leafs
@@ -0,0 +1,93 @@
+"""D14 layout: bin/{subject}/{method}.go, libs in internal/, one go.mod."""
+from __future__ import annotations
+
+import unittest
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+class BinLayoutTest(unittest.TestCase):
+    def test_brain_search_shebang_exists(self) -> None:
+        p = ROOT / "bin" / "brain" / "search.go"
+        self.assertTrue(p.is_file(), "missing bin/brain/search.go")
+        first = p.read_text().splitlines()[0]
+        self.assertTrue(
+            first.startswith("//usr/bin/env go run"),
+            f"shebang first line, got {first!r}",
+        )
+
+    def test_no_nested_go_mod_under_bin(self) -> None:
+        nested = list((ROOT / "bin").rglob("go.mod"))
+        self.assertEqual(nested, [], f"nested go.mod files: {nested}")
+
+    def test_rank_lives_in_internal_brain(self) -> None:
+        self.assertTrue(
+            (ROOT / "internal" / "brain" / "rank" / "rank.go").is_file(),
+            "ranking must live in internal/brain/rank (cgo-free)",
+        )
+        self.assertFalse(
+            (ROOT / "bin" / "kbsearch").exists(),
+            "bin/kbsearch nested module must be gone",
+        )
+
+    def test_no_main_go_under_bin_brain(self) -> None:
+        main = ROOT / "bin" / "brain" / "main.go"
+        self.assertFalse(main.exists(), "bin/brain/main.go is not a method")
+
+    def test_chats_methods_are_shebangs_not_main(self) -> None:
+        chats = ROOT / "bin" / "chats"
+        self.assertFalse(
+            (chats / "main.go").exists(),
+            "bin/chats/main.go is a dispatcher, not a method",
+        )
+        self.assertFalse(
+            (chats / "index_cmd.go").exists(),
+            "chats index is a brain write hiding under the wrong subject",
+        )
+        for method in ("sync.go", "import.go", "facts.go", "apply.go"):
+            p = chats / method
+            self.assertTrue(p.is_file(), f"missing bin/chats/{method}")
+            first = p.read_text().splitlines()[0]
+            self.assertTrue(
+                first.startswith("//usr/bin/env go run"),
+                f"{method} shebang, got {first!r}",
+            )
+
+    def test_chats_lib_lives_in_internal(self) -> None:
+        self.assertTrue(
+            (ROOT / "internal" / "chats" / "linkedin.go").is_file(),
+            "LinkedIn parser must live in internal/chats",
+        )
+        self.assertFalse(
+            (ROOT / "bin" / "chats" / "linkedin.go").exists(),
+            "parser must not stay under bin/chats as a second main",
+        )
+
+    def _assert_shebang(self, rel: str) -> None:
+        p = ROOT / rel
+        self.assertTrue(p.is_file(), f"missing {rel}")
+        first = p.read_text().splitlines()[0]
+        self.assertTrue(
+            first.startswith("//usr/bin/env go run"),
+            f"{rel} shebang, got {first!r}",
+        )
+
+    def test_brain_methods_are_shebangs(self) -> None:
+        for method in ("index.go", "get.go", "stats.go", "eval.go", "watch.go"):
+            self._assert_shebang(f"bin/brain/{method}")
+
+    def test_mail_import_is_shebang_not_brain_write(self) -> None:
+        self._assert_shebang("bin/mail/import.go")
+        index_mail = (ROOT / "bin" / "mail" / "index_mail").read_text()
+        self.assertIn(
+            "bin/brain/index.go",
+            index_mail,
+            "index_mail must point at bin/brain/index.go",
+        )
+
+    def test_markdown_import_is_shebang(self) -> None:
+        self._assert_shebang("bin/markdown/import.go")
+
+    def test_postgres_query_is_shebang(self) -> None:
+        self._assert_shebang("bin/postgres/query.go")
@@ -0,0 +1,46 @@
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import crmfacts  # noqa: E402
+
+FM = """\
+schema: 2
+meta:
+  title: x
+orgs:
+- id: produktor
+  label: ProProdukt SL / produktor.io
+  kind: own
+  period: 2006–present
+  website: https://produktor.io
+- id: dyvenia
+  label: Dyvenia
+  kind: employer
+  period: 2023–2025
+clients:
+- name: One
+- name: Two
+timeline:
+- start: 2001
+"""
+
+
+class CorpusOrgsTest(unittest.TestCase):
+    def test_parses_label_kind_period(self):
+        orgs = crmfacts.corpus_orgs(FM)
+        self.assertEqual(orgs["produktor"]["label"], "ProProdukt SL / produktor.io")
+        self.assertEqual(orgs["produktor"]["kind"], "own")
+        self.assertEqual(orgs["dyvenia"]["kind"], "employer")
+
+    def test_does_not_leak_clients_into_orgs(self):
+        orgs = crmfacts.corpus_orgs(FM)
+        self.assertNotIn("One", orgs)
+        self.assertNotIn("Two", orgs)
+        self.assertNotIn("timeline", orgs)
+
+
+if __name__ == "__main__":
+    unittest.main()
@@ -0,0 +1,71 @@
+import os
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import kblib  # noqa: E402
+import gitimport  # noqa: E402
+
+SAMPLE = (
+    "\x1e" + "a1b2c3d" + "\x1f" + "Ada Lovelace" + "\x1f" + "ada@example.com"
+    + "\x1f" + "2026-08-10T12:00:00+01:00" + "\x1f" + "feat: first commit"
+    + "\n\nREADME.md\nsrc/main.c\n"
+)
+
+COMMIT_PERSON_SCHEMA = (
+    "CREATE NODE TABLE IF NOT EXISTS Commit (id STRING, repo STRING, subject STRING, "
+    "author STRING, email STRING, date STRING, PRIMARY KEY(id))"
+)
+PERSON_SCHEMA = (
+    "CREATE NODE TABLE IF NOT EXISTS Person (id STRING, name STRING, email STRING, PRIMARY KEY(id))"
+)
+HAS_VERSION_SCHEMA = "CREATE REL TABLE IF NOT EXISTS HAS_VERSION (FROM File TO Commit)"
+AUTHORED_SCHEMA = "CREATE REL TABLE IF NOT EXISTS AUTHORED (FROM Commit TO Person)"
+
+
+class GitGraphTest(unittest.TestCase):
+    def setUp(self):
+        self.dir = tempfile.mkdtemp()
+        self.dbpath = os.path.join(self.dir, "kb.lbug")
+        self.db, self.conn = kblib.connect(self.dbpath, read_only=False)
+        kblib.init_schema(self.conn)
+        self.conn.execute(COMMIT_PERSON_SCHEMA)
+        self.conn.execute(PERSON_SCHEMA)
+        self.conn.execute(HAS_VERSION_SCHEMA)
+        self.conn.execute(AUTHORED_SCHEMA)
+
+    def tearDown(self):
+        self.conn.close()
+        self.db.close()
+
+    def test_index_commits_creates_nodes_and_edges(self):
+        cs = gitimport.parse_log(SAMPLE)
+        gitimport.index_commits(self.conn, cs, "sample-repo")
+        rp = self.conn.execute("MATCH (p:Person) RETURN p.name, p.email").get_all()
+        self.assertEqual([tuple(r) for r in rp], [("Ada Lovelace", "ada@example.com")])
+        rc = self.conn.execute("MATCH (c:Commit) RETURN c.id, c.repo").get_all()
+        self.assertEqual(len(rc), 1)
+        self.assertEqual(rc[0][1], "sample-repo")
+        # File -[:HAS_VERSION]-> Commit -[:AUTHORED]-> Person
+        rf = self.conn.execute(
+            "MATCH (f:File)-[:HAS_VERSION]->(c:Commit)-[:AUTHORED]->(p:Person) "
+            "RETURN f.path, c.id, p.email").get_all()
+        paths = sorted(r[0] for r in rf)
+        self.assertEqual(paths, ["README.md", "src/main.c"])
+        self.assertTrue(all(r[2] == "ada@example.com" for r in rf))
+
+    def test_index_commits_idempotent(self):
+        cs = gitimport.parse_log(SAMPLE)
+        gitimport.index_commits(self.conn, cs, "sample-repo")
+        gitimport.index_commits(self.conn, cs, "sample-repo")
+        n = self.conn.execute("MATCH (c:Commit) RETURN count(*)").get_all()[0][0]
+        self.assertEqual(n, 1)
+        p = self.conn.execute("MATCH (p:Person) RETURN count(*)").get_all()[0][0]
+        self.assertEqual(p, 1)
+
+
+if __name__ == "__main__":
+    unittest.main()
@@ -0,0 +1,57 @@
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import gitimport  # noqa: E402
+
+SAMPLE = (
+    "\x1e" + "a1b2c3d" + "\x1f" + "Ada Lovelace" + "\x1f" + "ada@example.com"
+    + "\x1f" + "2026-08-10T12:00:00+01:00" + "\x1f" + "feat: first commit"
+    + "\n\nREADME.md\nsrc/main.c\n"
+    + "\x1e" + "e4f5a6b" + "\x1f" + "Bob Babbage" + "\x1f" + "bob@example.com"
+    + "\x1f" + "2026-08-11T09:30:00+01:00" + "\x1f" + "fix: typo"
+    + "\n\ndocs/notes.md"
+)
+
+
+class GitparseTest(unittest.TestCase):
+    def test_parses_records(self):
+        cs = gitimport.parse_log(SAMPLE)
+        self.assertEqual(len(cs), 2)
+
+    def test_parses_commit_fields(self):
+        cs = gitimport.parse_log(SAMPLE)
+        c = cs[0]
+        self.assertEqual(c.sha, "a1b2c3d")
+        self.assertEqual(c.author, "Ada Lovelace")
+        self.assertEqual(c.email, "ada@example.com")
+        self.assertEqual(c.date, "2026-08-10T12:00:00+01:00")
+        self.assertEqual(c.subject, "feat: first commit")
+
+    def test_parses_changed_files(self):
+        cs = gitimport.parse_log(SAMPLE)
+        self.assertEqual(cs[0].files, ["README.md", "src/main.c"])
+        self.assertEqual(cs[1].files, ["docs/notes.md"])
+
+    def test_ignores_empty(self):
+        self.assertEqual(gitimport.parse_log(""), [])
+
+    def test_skip_malformed_record(self):
+        self.assertEqual(gitimport.parse_log("\x1eweird\x1e"), [])
+
+    def test_commit_leaf_shape(self):
+        leafs = gitimport.commits_to_leafs(gitimport.parse_log(SAMPLE), "sample-repo")
+        self.assertEqual(len(leafs), 2)
+        lf = leafs[0]
+        self.assertEqual(lf["type"], "commit")
+        self.assertEqual(lf["repo"], "sample-repo")
+        self.assertEqual(lf["source"], "sample-repo@a1b2c3d")
+        self.assertIn("Ada Lovelace", lf["text"])
+        self.assertIn("README.md", lf["related"])
+        self.assertIn("feat: first commit", lf["heading"])
+
+
+if __name__ == "__main__":
+    unittest.main()
@@ -35,7 +35,7 @@ class KblibTest(unittest.TestCase):
                          confidence="confirmed", source="s", source_rev="r1",
                          how="test", loc="/tmp", type_="reference",
                          embedding=make_emb(1.0))
-        kblib.create_fts_and_vector(self.conn, force=True)
+        kblib.ensure_indexes(self.conn)
        hits = kblib.query_fts(self.conn, "fox", 5)
        self.assertEqual(len(hits), 1)
        self.assertEqual(hits[0]["root"], "info")
@@ -49,12 +49,42 @@ class KblibTest(unittest.TestCase):
                          confidence="confirmed", source="s", source_rev="r1",
                          how="test", loc="/tmp", type_="reference",
                          embedding=make_emb(0.0))
-        kblib.create_fts_and_vector(self.conn, force=True)
+        kblib.ensure_indexes(self.conn)
        result = kblib.hybrid_search(self.conn, make_emb(1.0), [], 5)
        self.assertTrue(result)
        self.assertIn("rrf", result[0])
        self.assertEqual(result[0]["text"], "the quick brown fox")

+    def test_upsert_keeps_hnsw_queryable(self):
+        """Upsert while HNSW exists must not kill vector search."""
+        kblib.upsert_leaf(self.conn, text="seed leaf", root="info",
+                          confidence="confirmed", source="s", source_rev="r1",
+                          how="test", loc="/tmp", type_="reference",
+                          embedding=make_emb(0.2))
+        kblib.ensure_indexes(self.conn)
+        self.assertIn("Leaf_vec", kblib.leaf_index_names(self.conn))
+        kblib.upsert_leaf(self.conn, text="added after index", root="facts",
+                          confidence="confirmed", source="a.md x b.md",
+                          source_rev="r1", how="test", loc="/tmp", type_="fact",
+                          embedding=make_emb(0.9))
+        hits = kblib.query_vector(self.conn, make_emb(0.9), 5)
+        self.assertTrue(hits)
+        self.assertIn("Leaf_vec", kblib.leaf_index_names(self.conn))
+
+    def test_drop_vector_then_create_raises_clear_error(self):
+        """DROP INDEX leaves ghost catalog; create_fts_and_vector must raise."""
+        kblib.upsert_leaf(self.conn, text="seed", root="info",
+                          confidence="confirmed", source="s", source_rev="r1",
+                          how="test", loc="/tmp", type_="reference",
+                          embedding=make_emb(0.1))
+        kblib.ensure_indexes(self.conn)
+        self.conn.execute("DROP INDEX IF EXISTS Leaf.Leaf_vec")
+        with self.assertRaises(RuntimeError) as ctx:
+            kblib.create_fts_and_vector(self.conn, force=True)
+        msg = str(ctx.exception)
+        self.assertIn("CREATE_VECTOR_INDEX failed", msg)
+        self.assertIn("--rebuild", msg)
+
    def test_stats_counts_roots(self):
        kblib.upsert_leaf(self.conn, text="a fact leaf", root="facts",
                          confidence="confirmed", source="s", source_rev="r1",
@@ -70,4 +100,4 @@ class KblibTest(unittest.TestCase):


 if __name__ == "__main__":
-    unittest.main()
+    unittest.main()
@@ -0,0 +1,123 @@
+import io
+import os
+import sys
+import unittest
+import zipfile
+from pathlib import Path
+
+sys.path.insert(0, os.path.dirname(__file__))
+
+from mailconv import (  # noqa: E402
+    clean_email_address,
+    html_to_markdown,
+    is_convertible,
+    normalize_markdown,
+    split_zip_members,
+    subject_to_filename,
+    zip_extract_safe,
+)
+from mailconv import _unwrap_tables  # noqa: E402
+
+
+class TestMailConv(unittest.TestCase):
+    def test_clean_email_address(self):
+        self.assertEqual(clean_email_address('"Ben Baker" <bb@teks.com>'), "bb@teks.com")
+        self.assertEqual(clean_email_address("eslider@gmail.com"), "eslider@gmail.com")
+        self.assertEqual(clean_email_address("<a@b.c>"), "a@b.c")
+
+    def test_subject_to_filename(self):
+        self.assertEqual(subject_to_filename("Your receipt #2422"), "Your_receipt_2422")
+        self.assertEqual(subject_to_filename("a/b\\c:d*e"), "abcde")
+        self.assertEqual(subject_to_filename("   "), "untitled")
+
+    def test_html_to_markdown(self):
+        out = html_to_markdown("<html><body><h1>Hi</h1><p>Some <b>bold</b> text.</p></body></html>")
+        self.assertIn("Hi", out)
+        self.assertIn("**bold**", out)
+
+    def test_html_strip_fallback(self):
+        from mailconv import strip_html
+        self.assertEqual(strip_html("<p>a</p><p>b</p>"), "a\n\nb")
+
+    def test_flatten_layout_tables(self):
+        html = ("<table><tr>"
+                + "".join(f"<td>spacer{i}</td>" for i in range(12))
+                + "</tr></table>"
+                + "<p>real</p>"
+                + "<table><tr><td>a</td><td>b</td></tr></table>")
+        out = _unwrap_tables(html)
+        # tables unwrapped into pipe text; no <td> left; content preserved
+        self.assertNotIn("<td>spacer0</td>", out)
+        self.assertIn("spacer0 | spacer1", out)
+        self.assertIn("a | b", out)
+        self.assertIn("real", out)
+
+    def test_html_to_markdown_layout_clean(self):
+        html = "<table><tr>" + "".join(f"<td>x{i}</td>" for i in range(12)) + "</tr></table><h1>Hi</h1>"
+        out = html_to_markdown(html)
+        self.assertIn("Hi", out)
+        self.assertNotIn("| ---", out)
+
+    def test_normalize_markdown_removes_nul(self):
+        self.assertEqual(normalize_markdown("Z0\x00A\x00Y\x00B"), "Z0AYB")
+        self.assertEqual(normalize_markdown("a\n\n\n\nb"), "a\n\nb")
+
+    def test_normalize_strips_email_noise(self):
+        noisy = "\ufeffa\u200b\u034f\u00ad\u2007\u2002 b\u200a c\u2008"
+        out = normalize_markdown(noisy)
+        self.assertNotIn("\u200b", out)
+        self.assertNotIn("\ufeff", out)
+        self.assertNotIn("\u034f", out)
+        self.assertIn("a b c", out)
+
+    def test_split_zip_members(self):
+        p = Path(self._mk_zip(["a.txt", "sub/b.txt"]))
+        self.assertEqual(split_zip_members(p), ["a.txt", "sub/b.txt"])
+
+    def test_zip_extract_safe(self):
+        zip_path = self._mk_zip(["a.txt", "dir/b.txt"])
+        dest = Path(self._tmp("x"))
+        files = zip_extract_safe(zip_path, dest)
+        self.assertEqual(len(files), 2)
+        self.assertTrue((dest / "a.txt").exists())
+        self.assertTrue((dest / "dir" / "b.txt").exists())
+
+    def test_zip_extract_safe_blocks_traversal(self):
+        # member "../evil.txt" must not escape dest
+        zip_path = Path(self._tmp("evil.zip"))
+        with zipfile.ZipFile(zip_path, "w") as zf:
+            zf.writestr("../evil.txt", "boom")
+        dest = Path(self._tmp("out"))
+        files = zip_extract_safe(zip_path, dest)
+        self.assertEqual(files, [])
+        self.assertFalse((dest.parent / "evil.txt").exists())
+
+    def test_is_convertible(self):
+        self.assertTrue(is_convertible(".pdf"))
+        self.assertTrue(is_convertible(".zip"))
+        self.assertTrue(is_convertible(".docx"))
+        self.assertTrue(is_convertible(".TXT"))
+        self.assertFalse(is_convertible(".exe"))
+        self.assertFalse(is_convertible(".unknown"))
+
+    def _mk_zip(self, members):
+        zpath = Path(self._tmp("arc.zip"))
+        with zipfile.ZipFile(zpath, "w") as zf:
+            for m in members:
+                zf.writestr(m, "content")
+        return str(zpath)
+
+    def _tmp(self, name):
+        d = self.__class__._td
+        p = Path(d) / name
+        p.parent.mkdir(parents=True, exist_ok=True)
+        return str(p)
+
+    @classmethod
+    def setUpClass(cls):
+        import tempfile
+        cls._td = tempfile.mkdtemp(prefix="mailconv_test_")
+
+
+if __name__ == "__main__":
+    unittest.main()
@@ -0,0 +1,46 @@
+"""Mail markdown → leafs (no Ladybug). Brain index --with-mail uses this."""
+from __future__ import annotations
+
+import json
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+import mailleafs  # noqa: E402
+
+
+class MailLeafsTest(unittest.TestCase):
+    def test_message_md_becomes_info_leaf(self) -> None:
+        root = Path(tempfile.mkdtemp())
+        msg = root / "inbox" / "alice-1"
+        msg.mkdir(parents=True)
+        (msg / "message.json").write_text(
+            json.dumps({"receivedDate": "2026-01-15T10:00:00Z", "subject": "Hello"}),
+            encoding="utf-8",
+        )
+        (msg / "message.md").write_text(
+            "---\nroot: info\n---\n\n# Hello\n\nFrom Alice to Bob.\n",
+            encoding="utf-8",
+        )
+        leafs = mailleafs.from_mail_root(root)
+        self.assertEqual(len(leafs), 1)
+        self.assertIn("Alice", leafs[0]["text"])
+        self.assertTrue(leafs[0]["source"].startswith("ooMail:"))
+        self.assertEqual(leafs[0]["how"], "mail/import")
+
+    def test_since_filters_by_message_json_date(self) -> None:
+        root = Path(tempfile.mkdtemp())
+        for name, day in (("old", "2025-01-01"), ("new", "2026-06-01")):
+            d = root / "inbox" / name
+            d.mkdir(parents=True)
+            (d / "message.json").write_text(
+                json.dumps({"receivedDate": f"{day}T00:00:00Z"}),
+                encoding="utf-8",
+            )
+            (d / "message.md").write_text(f"# {name}\n\nbody\n", encoding="utf-8")
+        leafs = mailleafs.from_mail_root(root, since="2026-01-01")
+        self.assertEqual(len(leafs), 1)
+        self.assertIn("new", leafs[0]["text"])
@@ -0,0 +1,57 @@
+"""Published docs must match live commands (Gitea SoT, brain/search, no fake --hop)."""
+from __future__ import annotations
+
+import re
+import unittest
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[2]
+
+
+class PublishedDocsTest(unittest.TestCase):
+    def test_readme_points_issues_at_gitea(self) -> None:
+        text = (ROOT / "README.md").read_text()
+        self.assertIn(
+            "https://git.produktor.io/eSlider/2dph/issues",
+            text,
+            "README must point issues at Gitea",
+        )
+
+    def test_plan_d15_names_gitea_origin(self) -> None:
+        text = (ROOT / "PLAN.md").read_text()
+        self.assertIn("D15", text)
+        self.assertIn("git.produktor.io/eSlider/2dph", text)
+
+    def test_readme_primary_search_is_brain(self) -> None:
+        text = (ROOT / "README.md").read_text()
+        self.assertIn(
+            "bin/brain/search.go",
+            text,
+            "README deduction search must name bin/brain/search.go",
+        )
+
+    def test_readme_index_is_brain_not_index_mail(self) -> None:
+        text = (ROOT / "README.md").read_text()
+        self.assertIn("bin/brain/index.go", text)
+        self.assertNotIn(
+            "bin/mail/index_mail",
+            text,
+            "mail index is a brain write; README must name bin/brain/index.go",
+        )
+
+    def test_docs_do_not_claim_hop_walks(self) -> None:
+        paths = [
+            ROOT / "README.md",
+            ROOT / "docs" / "design.md",
+            ROOT / "skills" / "kb-search" / "SKILL.md",
+            ROOT / "skills" / "diataxis-docs" / "SKILL.md",
+        ]
+        # Command-style `--hop 1` / `--hop N` plus follow/walk = the old lie.
+        # Honest "not implemented" notes must not match.
+        lie = re.compile(r"--hop (?:N|1).*(?:follow|walk)", re.I | re.S)
+        for path in paths:
+            text = path.read_text()
+            self.assertIsNone(
+                lie.search(text),
+                f"{path.relative_to(ROOT)} still claims --hop walks the graph",
+            )
@@ -76,7 +76,7 @@ class PhiGuard(unittest.TestCase):

    def test_plain_technical_query_passes(self):
        self.assertIsNone(ws.phi_reason("Pflegegrad SGB XI Einstufung"))
-        self.assertIsNone(ws.phi_reason("site:ticket.detective.de Toureffizienz"))
+        self.assertIsNone(ws.phi_reason("site:example.com technical query"))

    def test_long_digit_run_is_refused(self):
        self.assertIsNotNone(ws.phi_reason("Kunde 4711220385 Adresse"))
@@ -0,0 +1,105 @@
+// Package watch polls corpus directories for changes and re-runs brain/index.
+//
+// Port of the former bin/kb-watch bash script to an importable, testable Go
+// package. Polls file mtimes (no inotify deps); cheap and reliable.
+package watch
+
+import (
+	"log"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// Options controls the polling loop. Zero value uses defaults.
+type Options struct {
+	Dirs     []string
+	Interval time.Duration
+	// IndexCmd is the index command template. %s is replaced by the repo
+	// root (from KB_ROOT). Defaults to `python3 <root>/bin/kb/index --with-mail`.
+	IndexCmd string
+}
+
+// Run blocks forever polling Dirs (defaults: KB_WATCH_DIRS or /corpus) every
+// Interval (default 30s) and re-indexing when files change. KB_ROOT names the
+// repo root used to locate bin/kb/index.
+func Run(args []string) {
+	opts := fromEnv(args)
+	root, _ := os.Getwd()
+	if r := os.Getenv("KB_ROOT"); r != "" {
+		root = r
+	}
+	log.Printf("watch: dirs=%v interval=%s root=%s", opts.Dirs, opts.Interval, root)
+	var last string
+	for {
+		if flag := Stamp(opts.Dirs); flag != "" && flag != last {
+			last = flag
+			reindex(opts.IndexCmd, root)
+		}
+		time.Sleep(opts.Interval)
+	}
+}
+
+func fromEnv(args []string) Options {
+	opts := Options{Interval: 30 * time.Second}
+	if raw := os.Getenv("KB_WATCH_INTERVAL"); raw != "" {
+		if n, err := strconv.Atoi(raw); err == nil && n > 0 {
+			opts.Interval = time.Duration(n) * time.Second
+		}
+	}
+	defDirs := "/corpus"
+	if raw := os.Getenv("KB_WATCH_DIRS"); raw != "" {
+		defDirs = raw
+	}
+	if len(args) > 0 {
+		opts.Dirs = args
+	} else {
+		for _, d := range strings.Split(defDirs, " ") {
+			if d != "" {
+				opts.Dirs = append(opts.Dirs, d)
+			}
+		}
+	}
+	pys := os.Getenv("KB_PY")
+	if pys == "" {
+		pys = "python3"
+	}
+	opts.IndexCmd = pys + " <root>/bin/kb/index --with-mail"
+	return opts
+}
+
+// Stamp returns a rolling fingerprint (newest mtime under dirs) that changes
+// whenever any corpus file is touched. Empty when no files found.
+func Stamp(dirs []string) string {
+	var newest time.Time
+	for _, dir := range dirs {
+		_ = filepath.WalkDir(dir, func(path string, _ os.DirEntry, err error) error {
+			if err != nil {
+				return nil
+			}
+			if info, e := os.Stat(path); e == nil && info.ModTime().After(newest) {
+				newest = info.ModTime()
+			}
+			return nil
+		})
+	}
+	if newest.IsZero() {
+		return ""
+	}
+	return strconv.FormatInt(newest.UnixNano(), 10)
+}
+
+func reindex(template, root string) {
+	cmd := strings.ReplaceAll(template, "<root>", root)
+	parts := strings.Fields(cmd)
+	c := exec.Command(parts[0], parts[1:]...)
+	out, err := c.CombinedOutput()
+	if err != nil {
+		log.Printf("watch: index failed: %v\n%s", err, out)
+	} else {
+		log.Printf("watch: re-indexed")
+	}
+}
@@ -0,0 +1,53 @@
+package watch
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func TestStampChangesWhenFileTouched(t *testing.T) {
+	dir := t.TempDir()
+	a := filepath.Join(dir, "a.md")
+	if err := os.WriteFile(a, []byte("x"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	s1 := Stamp([]string{dir})
+	if s1 == "" {
+		t.Fatal("stamp empty for a dir with a file")
+	}
+	time.Sleep(10 * time.Millisecond)
+	if err := os.WriteFile(a, []byte("y"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if s2 := Stamp([]string{dir}); s2 == s1 {
+		t.Fatal("stamp did not change after the file was modified")
+	}
+}
+
+func TestStampEmptyForMissingDir(t *testing.T) {
+	if s := Stamp([]string{filepath.Join(t.TempDir(), "nope")}); s != "" {
+		t.Fatalf("stamp = %q, want empty for missing dir", s)
+	}
+}
+
+func TestFromEnvDefaults(t *testing.T) {
+	t.Setenv("KB_WATCH_INTERVAL", "")
+	t.Setenv("KB_WATCH_DIRS", "")
+	t.Setenv("KB_PY", "")
+	opts := fromEnv(nil)
+	if len(opts.Dirs) == 0 || opts.Dirs[0] != "/corpus" {
+		t.Fatalf("default dirs = %v, want [/corpus]", opts.Dirs)
+	}
+	if opts.Interval != 30*time.Second {
+		t.Fatalf("default interval = %s, want 30s", opts.Interval)
+	}
+	if !strings.Contains(opts.IndexCmd, "kb/index") {
+		t.Fatalf("default index cmd = %q, want kb/index", opts.IndexCmd)
+	}
+	if !strings.Contains(opts.IndexCmd, "--with-mail") {
+		t.Fatalf("default index cmd must include --with-mail, got %q", opts.IndexCmd)
+	}
+}
@@ -25,7 +25,7 @@ import urllib.parse
 import urllib.request
 from pathlib import Path

-TOOLS = Path(__file__).resolve().parents[1].parent / "tools"
+TOOLS = Path(__file__).resolve().parents[1] / "tools"
 sys.path.insert(0, str(TOOLS))
 sys.path.insert(0, str(TOOLS / "web-search"))

@@ -6,5 +6,10 @@ Brain/ops/eSlider stack. Facts need proof or they are

 - [PLAN.md](../PLAN.md) — decisions, execution order, open questions (v2)
 - [design](design.md) — schema, deduction model, sources
+- [Gitea issues](https://git.produktor.io/eSlider/2dph/issues) — work board (origin)

-Published docs live here and mirror the project state.
+Search: `bin/brain/search.go "query"` (HTTP: `bin/brain/serve.go` —
+`/health` `/search` `/get` `/stats` `/audit` `/ingest`). `--hop` is
+not a walk; the flag errors until File/FROM_FILE edges exist.
+
+Published docs live here and mirror the project state.
@@ -0,0 +1,24 @@
+# Chat Import Pipeline
+
+Plan: https://git.produktor.io/eSlider/brain-chats-import/issues/1
+
+## Env vars (set in shell, never committed)
+
+```
+TELEGRAM_MCP_DIR
+TELEGRAM_API_ID / TELEGRAM_API_HASH / TELEGRAM_PHONE
+TELEGRAM_SESSION_STRING
+ONLYOFFICE_URL / ONLYOFFICE_USER / ONLYOFFICE_PASS
+OO_CLI                     (default: $HOME/go/bin/oo)
+```
+
+## Quick reference
+
+```
+./bin/chats/sync.go telegram --limit 100
+./bin/chats/import.go
+./bin/chats/facts.go
+./bin/chats/apply.go --dry-run
+```
+
+JSONL → markdown only. Brain ingest is `bin/brain/index.go` (not a `chats index`).
@@ -0,0 +1,33 @@
+# CRM association proof (oo CLI ↔ corpus)
+
+Proven with `oo` (eslider/go-onlyoffice) against the OnlyOffice portal
+(`office.produktor.io`). Portal CRM is the SSOT for company ↔ person ↔
+project associations; the corpus SoT (`eslider/cv/projects/knowledge-mesh-seed.yaml`)
+is the second, independent source. Facts that can be backed by both are
+written to the brain under `root=facts` by `bin/facts/crm`.
+
+## What was verified
+
+- Logical counts (portal MySQL): 1300 contacts = 897 persons + 404 companies,
+  198 projects, 998 deals, 939 project↔contact links.
+- Every client company linked to a project has ≥1 person underneath.
+- Every person `company_id` resolves to an existing company.
+- Corpus org list (9) maps 1:1 onto CRM companies:
+  ProProdukt SL / produktor.io, Dyvenia, Immowelt AG, WhereGroup,
+  Keynote SIGOS, D2S/SYSTEMS, GRID, Pack und Cup, Markets Platform.
+- 78 person↔company association facts written to the brain
+  (`how=crm-crosscheck`, `type=association`). Recall@5 in `bin/kb/eval` = 1.0.
+
+## Mistakes found
+
+| # | Mistake | Fix |
+|---|---------|-----|
+| 1 | Duplicate legal entity `GoldenRatio.Exchange` (contact 759) vs `Golden Ratio Exchange` (763); 3 deals (211, 287, 559) were linked to 759 | `oo contacts merge 759 763` — 763 kept, 759 removed, deal links re-pointed to 763 |
+| 2 | `env/`-wide: OnlyOffice creds file used wrong UX (user `eslider`, password with `$2` suffix) making `oo` auth fail | `.env` fixed to `eslider@gmail.com` + clean password; `.env` stays gitignored |
+
+## Gates after fix
+
+- `uv run python -m unittest discover -s bin/tools -t .` → 26 tests OK
+- `bin/facts/audit self` + `bin/facts/audit db` → ok
+- `bin/kb/eval` → recall@5 = 1.0
+- `go test ./...` (bin/server + bin/watch) → ok
@@ -17,14 +17,14 @@ as one consistent state.
 ## Deduction search

 ```
-bin/kb/search "question"
+bin/brain/search.go "question"
  1. facts root   — confirmed answers only   → return with evidence links
  2. info  root   — supporting narrative     → snippets, marked (not confirmed)
  3. web-search   — second independent source → upgrade hypothesis to confirmed
 ```

-`--hop N` follows graph edges (sibling leaves under a heading, owning file,
-`related:` files, vector-neighbour leaves) — the deduction walk.
+`--hop` is not implemented yet (needs File/FROM_FILE edges). The flag is an
+error; it is not a graph walk.

 ## Who / What / How / Where / When + evidence

@@ -0,0 +1,25 @@
+module github.com/eSlider/2dph
+
+go 1.26
+
+require (
+	github.com/LadybugDB/go-ladybug v0.17.0
+	github.com/arran4/golang-ical v0.3.5
+	github.com/chewxy/math32 v1.11.2
+	github.com/daulet/tokenizers v1.27.0
+	golang.org/x/text v0.40.0
+)
+
+require (
+	github.com/apache/arrow-go/v18 v18.6.0 // indirect
+	github.com/goccy/go-json v0.10.6 // indirect
+	github.com/google/flatbuffers v25.12.19+incompatible // indirect
+	github.com/google/uuid v1.6.0 // indirect
+	github.com/klauspost/compress v1.18.5 // indirect
+	github.com/klauspost/cpuid/v2 v2.3.0 // indirect
+	github.com/pierrec/lz4/v4 v4.1.26 // indirect
+	github.com/shopspring/decimal v1.4.0 // indirect
+	github.com/zeebo/xxh3 v1.1.0 // indirect
+	golang.org/x/exp v0.0.0-20260112195511-716be5621a96 // indirect
+	golang.org/x/sys v0.43.0 // indirect
+)
@@ -0,0 +1,50 @@
+github.com/LadybugDB/go-ladybug v0.17.0 h1:RXDbkBjrbRmLdEbhGl4CLOIEzSt09gbP0n9UbKDEfwI=
+github.com/LadybugDB/go-ladybug v0.17.0/go.mod h1:GeIXmE8XyF5TFS94NAuTag7vgCC+no/HTBMRA6Rd5Cs=
+github.com/andybalholm/brotli v1.2.1 h1:R+f5xP285VArJDRgowrfb9DqL18yVK0gKAW/F+eTWro=
+github.com/andybalholm/brotli v1.2.1/go.mod h1:rzTDkvFWvIrjDXZHkuS16NPggd91W3kUSvPlQ1pLaKY=
+github.com/apache/arrow-go/v18 v18.6.0 h1:GX/Jyd3R7mCLiECAwY9FWbbaYblie2WXBSz4Sw8fNpM=
+github.com/apache/arrow-go/v18 v18.6.0/go.mod h1:gm3MiPpY82fLYK5VKPB3WoJbsiLVDfT7flD5/vHReKw=
+github.com/apache/thrift v0.22.0 h1:r7mTJdj51TMDe6RtcmNdQxgn9XcyfGDOzegMDRg47uc=
+github.com/apache/thrift v0.22.0/go.mod h1:1e7J/O1Ae6ZQMTYdy9xa3w9k+XHWPfRvdPyJeynQ+/g=
+github.com/arran4/golang-ical v0.3.5 h1:bbz6ld4dC+MmCKiFfOd6SkmIGnhNMBACZ485ULh7p9A=
+github.com/arran4/golang-ical v0.3.5/go.mod h1:OnguFgjN0Hmx8jzpmWcC+AkHio94ujmLHKoaef7xQh8=
+github.com/chewxy/math32 v1.11.2 h1:IufN08Zwr1NKuWfY+4Tz55BcwKmyKKNdOP7KtumehnM=
+github.com/chewxy/math32 v1.11.2/go.mod h1:dOB2rcuFrCn6UHrze36WSLVPKtzPMRAQvBvUwkSsLqs=
+github.com/daulet/tokenizers v1.27.0 h1:MmFYAEDFz69s/nNQfHg59DWqHz3v94m99kEZ/JbL+s4=
+github.com/daulet/tokenizers v1.27.0/go.mod h1:YjFY1o1HGMyWkQgbXJDghhvke/yFDp2vGdIO2hYs4MQ=
+github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc h1:U9qPSI2PIWSS1VwoXQT9A3Wy9MM3WgvqSxFWenqJduM=
+github.com/davecgh/go-spew v1.1.2-0.20180830191138-d8f796af33cc/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+github.com/goccy/go-json v0.10.6 h1:p8HrPJzOakx/mn/bQtjgNjdTcN+/S6FcG2CTtQOrHVU=
+github.com/goccy/go-json v0.10.6/go.mod h1:oq7eo15ShAhp70Anwd5lgX2pLfOS3QCiwU/PULtXL6M=
+github.com/google/flatbuffers v25.12.19+incompatible h1:haMV2JRRJCe1998HeW/p0X9UaMTK6SDo0ffLn2+DbLs=
+github.com/google/flatbuffers v25.12.19+incompatible/go.mod h1:1AeVuKshWv4vARoZatz6mlQ0JxURH0Kv5+zNeJKJCa8=
+github.com/google/go-cmp v0.6.0 h1:ofyhxvXcZhMsU5ulbFiLKl/XBFqE1GSq7atu8tAmTRI=
+github.com/google/go-cmp v0.6.0/go.mod h1:17dUlkBOakJ0+DkrSSNjCkIjxS6bF9zb3elmeNGIjoY=
+github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
+github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/klauspost/compress v1.18.5 h1:/h1gH5Ce+VWNLSWqPzOVn6XBO+vJbCNGvjoaGBFW2IE=
+github.com/klauspost/compress v1.18.5/go.mod h1:cwPg85FWrGar70rWktvGQj8/hthj3wpl0PGDogxkrSQ=
+github.com/klauspost/cpuid/v2 v2.3.0 h1:S4CRMLnYUhGeDFDqkGriYKdfoFlDnMtqTiI/sFzhA9Y=
+github.com/klauspost/cpuid/v2 v2.3.0/go.mod h1:hqwkgyIinND0mEev00jJYCxPNVRVXFQeu1XKlok6oO0=
+github.com/pierrec/lz4/v4 v4.1.26 h1:GrpZw1gZttORinvzBdXPUXATeqlJjqUG/D87TKMnhjY=
+github.com/pierrec/lz4/v4 v4.1.26/go.mod h1:EoQMVJgeeEOMsCqCzqFm2O0cJvljX2nGZjcRIPL34O4=
+github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 h1:Jamvg5psRIccs7FGNTlIRMkT8wgtp5eCXdBlqhYGL6U=
+github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/shopspring/decimal v1.4.0 h1:bxl37RwXBklmTi0C79JfXCEBD1cqqHt0bbgBAGFp81k=
+github.com/shopspring/decimal v1.4.0/go.mod h1:gawqmDU56v4yIKSwfBSFip1HdCCXN8/+DMd9qYNcwME=
+github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
+github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
+github.com/zeebo/assert v1.3.0 h1:g7C04CbJuIDKNPFHmsk4hwZDO5O+kntRxzaUoNXj+IQ=
+github.com/zeebo/assert v1.3.0/go.mod h1:Pq9JiuJQpG8JLJdtkwrJESF0Foym2/D9XMU5ciN/wJ0=
+github.com/zeebo/xxh3 v1.1.0 h1:s7DLGDK45Dyfg7++yxI0khrfwq9661w9EN78eP/UZVs=
+github.com/zeebo/xxh3 v1.1.0/go.mod h1:IisAie1LELR4xhVinxWS5+zf1lA4p0MW4T+w+W07F5s=
+golang.org/x/exp v0.0.0-20260112195511-716be5621a96 h1:Z/6YuSHTLOHfNFdb8zVZomZr7cqNgTJvA8+Qz75D8gU=
+golang.org/x/exp v0.0.0-20260112195511-716be5621a96/go.mod h1:nzimsREAkjBCIEFtHiYkrJyT+2uy9YZJB7H1k68CXZU=
+golang.org/x/sys v0.43.0 h1:Rlag2XtaFTxp19wS8MXlJwTvoh8ArU6ezoyFsMyCTNI=
+golang.org/x/sys v0.43.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
+golang.org/x/text v0.40.0 h1:Ub2Z6/xjgF1WrYQz2nuITOEegKFtiIy+rieRJ5lHZKs=
+golang.org/x/text v0.40.0/go.mod h1:hpnzDAfGV753zIKo+wk3u1bVKCGPbrnF7+7LBF/UHVY=
+gonum.org/v1/gonum v0.17.0 h1:VbpOemQlsSMrYmn7T2OUvQ4dqxQXU+ouZFQsZOx50z4=
+gonum.org/v1/gonum v0.17.0/go.mod h1:El3tOrEuMpv2UdMrbNlKEh9vd86bmQ6vqIcDwxEOc1E=
+gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
+gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
@@ -0,0 +1,3 @@
+go 1.26
+
+use .
@@ -0,0 +1,100 @@
+//go:build cgo && system_ladybug
+
+package brain
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	lbug "github.com/LadybugDB/go-ladybug"
+)
+
+var (
+	db   *lbug.Database
+	conn *lbug.Connection
+)
+
+func repoRoot() string {
+	// Try KB_ROOT env, then walk up from binary
+	if v := os.Getenv("KB_ROOT"); v != "" {
+		return v
+	}
+	self, err := os.Executable()
+	if err == nil {
+		dir := filepath.Dir(self)
+		for i := 0; i < 5; i++ {
+			if _, err := os.Stat(filepath.Join(dir, "var")); err == nil {
+				return dir
+			}
+			if _, err := os.Stat(filepath.Join(dir, ".git")); err == nil {
+				return dir
+			}
+			parent := filepath.Dir(dir)
+			if parent == dir {
+				break
+			}
+			dir = parent
+		}
+	}
+	return "."
+}
+
+func dbPath() string {
+	return filepath.Join(repoRoot(), "var", "kb.lbug")
+}
+
+func openBrain() error {
+	return openWithSandbox(eps())
+}
+
+func openWithSandbox(epsv string) error {
+	cfg := lbug.DefaultSystemConfig()
+	cfg.MaxNumThreads = 8
+	cfg.BufferPoolSize = 1 << 30 // 1GB
+
+	var err error
+	db, err = lbug.OpenDatabase(dbPath(), cfg)
+	if err != nil {
+		return fmt.Errorf("OpenDatabase: %w", err)
+	}
+
+	conn, err = lbug.OpenConnection(db)
+	if err != nil {
+		closeBrain()
+		return fmt.Errorf("OpenConnection: %w", err)
+	}
+	// Session settings need a live connection; running this before
+	// OpenConnection dereferenced a nil *Connection.
+	if epsv != "" {
+		if strings.ContainsAny(epsv, "'\\") {
+			closeBrain()
+			return fmt.Errorf("SET STREAM_SANDBOX: invalid value")
+		}
+		if _, err := conn.Query("SET STREAM_SANDBOX = '" + epsv + "'"); err != nil {
+			closeBrain()
+			return fmt.Errorf("SET STREAM_SANDBOX: %w", err)
+		}
+	}
+	if _, err := conn.Query("LOAD EXTENSION FTS"); err != nil {
+		closeBrain()
+		return fmt.Errorf("LOAD EXTENSION FTS: %w", err)
+	}
+	if _, err := conn.Query("LOAD EXTENSION VECTOR"); err != nil {
+		closeBrain()
+		return fmt.Errorf("LOAD EXTENSION VECTOR: %w", err)
+	}
+	return nil
+}
+
+func closeBrain() {
+	if conn != nil {
+		conn.Close()
+		conn = nil
+	}
+	if db != nil {
+		db.Close()
+		db = nil
+	}
+}
@@ -0,0 +1,3 @@
+// Package brain is deduction search over Ladybug (FTS + HNSW).
+// Query/embed code that needs cgo lives behind the system_ladybug tag.
+package brain
@@ -0,0 +1,154 @@
+//go:build cgo && system_ladybug
+
+package brain
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+)
+
+// Ready opens the Ladybug file for the life of the serve process.
+func Ready() error {
+	return openBrain()
+}
+
+// HTTP is the in-process API used by bin/brain/serve.go.
+type HTTP struct{}
+
+func (HTTP) Search(_ context.Context, query string, limit int) ([]byte, error) {
+	hits, err := searchHits(query, "", "", limit)
+	if err != nil {
+		return nil, err
+	}
+	for i := range hits {
+		if hits[i].Text != "" {
+			runes := []rune(hits[i].Text)
+			if len(runes) > 280 {
+				runes = runes[:280]
+			}
+			hits[i].Snippet = string(runes)
+		}
+	}
+	var buf bytes.Buffer
+	enc := json.NewEncoder(&buf)
+	enc.SetEscapeHTML(false)
+	if err := enc.Encode(toJSONOut(hits, query, "")); err != nil {
+		return nil, err
+	}
+	return buf.Bytes(), nil
+}
+
+func (HTTP) Get(_ context.Context, id string, body bool) ([]byte, error) {
+	if conn == nil {
+		return nil, fmt.Errorf("brain not open")
+	}
+	stmt, err := conn.Prepare(
+		"MATCH (l:Leaf {id:$id}) RETURN l.id, l.text, l.root, l.confidence, l.source, l.type",
+	)
+	if err != nil {
+		return nil, err
+	}
+	defer stmt.Close()
+	res, err := conn.Execute(stmt, map[string]any{"id": id})
+	if err != nil {
+		return nil, err
+	}
+	if !res.HasNext() {
+		return nil, fmt.Errorf("no leaf %s", id)
+	}
+	row, err := res.Next()
+	if err != nil {
+		return nil, err
+	}
+	vals, err := row.GetAsSlice()
+	if err != nil || len(vals) < 6 {
+		return nil, fmt.Errorf("leaf row")
+	}
+	out := map[string]any{
+		"id":         fmt.Sprint(vals[0]),
+		"root":       fmt.Sprint(vals[2]),
+		"confidence": fmt.Sprint(vals[3]),
+		"source":     fmt.Sprint(vals[4]),
+		"type":       fmt.Sprint(vals[5]),
+	}
+	if body {
+		out["text"] = fmt.Sprint(vals[1])
+	}
+	return json.Marshal(out)
+}
+
+func (HTTP) Stats(context.Context) ([]byte, error) {
+	if conn == nil {
+		return nil, fmt.Errorf("brain not open")
+	}
+	res, err := conn.Query("MATCH (l:Leaf) RETURN l.root, count(*)")
+	if err != nil {
+		return nil, err
+	}
+	byRoot := map[string]int{}
+	total := 0
+	for res.HasNext() {
+		row, err := res.Next()
+		if err != nil {
+			return nil, err
+		}
+		vals, err := row.GetAsSlice()
+		if err != nil || len(vals) < 2 {
+			continue
+		}
+		n := int(asInt(vals[1]))
+		byRoot[fmt.Sprint(vals[0])] = n
+		total += n
+	}
+	return json.Marshal(map[string]any{"total": total, "by_root": byRoot, "db": dbPath()})
+}
+
+func (HTTP) Audit(context.Context) ([]byte, error) {
+	if conn == nil {
+		return nil, fmt.Errorf("brain not open")
+	}
+	res, err := conn.Query("MATCH (l:Leaf) RETURN l.root, l.confidence, count(*)")
+	if err != nil {
+		return nil, err
+	}
+	var rows []map[string]any
+	for res.HasNext() {
+		row, err := res.Next()
+		if err != nil {
+			return nil, err
+		}
+		vals, err := row.GetAsSlice()
+		if err != nil || len(vals) < 3 {
+			continue
+		}
+		rows = append(rows, map[string]any{
+			"root":       fmt.Sprint(vals[0]),
+			"confidence": fmt.Sprint(vals[1]),
+			"count":      asInt(vals[2]),
+		})
+	}
+	return json.Marshal(map[string]any{"status": "ok", "by_confidence": rows})
+}
+
+func (HTTP) Ingest(context.Context) ([]byte, error) {
+	return json.Marshal(map[string]any{
+		"mode":    "rebuild",
+		"command": "bin/brain/index.go --rebuild",
+		"add":     "v2",
+	})
+}
+
+func asInt(v any) int64 {
+	switch n := v.(type) {
+	case int64:
+		return n
+	case int:
+		return int64(n)
+	case float64:
+		return int64(n)
+	default:
+		return 0
+	}
+}
@@ -0,0 +1,238 @@
+//go:build cgo && system_ladybug
+
+// StaticModel wraps the potion-multilingual-128m embedding model.
+//
+// Mirrors model2vec.StaticModel: tokenizer (daulet Unigram) + safetensors matrix.
+// Embed(text) applies the same preprocessing: median_token_length pre-truncation,
+// add_special_tokens=false, drop unk (id=1), truncate to 512, mean pool, L2 normalize +1e-32.
+package brain
+
+import (
+	"encoding/json"
+	"fmt"
+	"io"
+	"math"
+	"os"
+	"path/filepath"
+	"sort"
+
+	"github.com/chewxy/math32"
+	"github.com/daulet/tokenizers"
+)
+
+type StaticModel struct {
+	tok       *tokenizers.Tokenizer
+	mat       []float32 // row-major: vocab_size x 128
+	medianLen int
+	vocabSize int
+	dim       int
+}
+
+func loadModel() (*StaticModel, error) {
+	dir, err := modelDir()
+	if err != nil {
+		return nil, err
+	}
+
+	tok, err := tokenizers.FromFile(filepath.Join(dir, "tokenizer.json"))
+	if err != nil {
+		return nil, fmt.Errorf("tokenizer: %w", err)
+	}
+
+	mat, vocabSize, dim, err := loadMatrix(filepath.Join(dir, "model.safetensors"))
+	if err != nil {
+		return nil, fmt.Errorf("safetensors: %w", err)
+	}
+
+	median := medianTokenLength(filepath.Join(dir, "tokenizer.json"))
+
+	return &StaticModel{
+		tok:       tok,
+		mat:       mat,
+		vocabSize: vocabSize,
+		dim:       dim,
+		medianLen: median,
+	}, nil
+}
+
+func (m *StaticModel) Close() error {
+	if m.tok != nil {
+		m.tok.Close()
+		m.tok = nil
+	}
+	return nil
+}
+
+func (m *StaticModel) Embed(text string) ([]float64, error) {
+	const maxLen = 512
+
+	if m.medianLen > 0 {
+		maxChars := maxLen * m.medianLen
+		runes := []rune(text)
+		if len(runes) > maxChars {
+			text = string(runes[:maxChars])
+		}
+	}
+
+	ids, _, err := m.tok.EncodeErr(text, false)
+	if err != nil {
+		return nil, fmt.Errorf("encode: %w", err)
+	}
+
+	filtered := make([]uint32, 0, len(ids))
+	for _, id := range ids {
+		if id != 1 {
+			filtered = append(filtered, id)
+		}
+		if len(filtered) >= maxLen {
+			break
+		}
+	}
+	if len(filtered) == 0 {
+		return make([]float64, m.dim), nil
+	}
+
+	acc := make([]float32, m.dim)
+	for _, id := range filtered {
+		if int(id) >= m.vocabSize {
+			continue
+		}
+		off := int(id) * m.dim
+		for d := 0; d < m.dim; d++ {
+			acc[d] += m.mat[off+d]
+		}
+	}
+	inv := 1.0 / float32(len(filtered))
+	for d := 0; d < m.dim; d++ {
+		acc[d] *= inv
+	}
+
+	var norm float32
+	for d := 0; d < m.dim; d++ {
+		norm += acc[d] * acc[d]
+	}
+	norm = math32.Sqrt(norm) + 1e-32
+	for d := 0; d < m.dim; d++ {
+		acc[d] /= norm
+	}
+
+	out := make([]float64, m.dim)
+	for d := 0; d < m.dim; d++ {
+		out[d] = float64(acc[d])
+	}
+	return out, nil
+}
+
+func medianTokenLength(tokenizerPath string) int {
+	data, err := os.ReadFile(tokenizerPath)
+	if err != nil {
+		return 0
+	}
+	var parsed struct {
+		Model struct {
+			Vocab [][]json.RawMessage `json:"vocab"`
+		} `json:"model"`
+	}
+	if err := json.Unmarshal(data, &parsed); err != nil {
+		return 0
+	}
+	vocab := parsed.Model.Vocab
+	if len(vocab) == 0 {
+		return 0
+	}
+	lengths := make([]int, 0, len(vocab))
+	for _, pair := range vocab {
+		if len(pair) < 1 {
+			continue
+		}
+		var tok string
+		if err := json.Unmarshal(pair[0], &tok); err != nil {
+			continue
+		}
+		lengths = append(lengths, len([]rune(tok)))
+	}
+	if len(lengths) == 0 {
+		return 0
+	}
+	sort.Ints(lengths)
+	return lengths[len(lengths)/2]
+}
+
+func loadMatrix(path string) ([]float32, int, int, error) {
+	f, err := os.Open(path)
+	if err != nil {
+		return nil, 0, 0, err
+	}
+	defer f.Close()
+
+	var hdrLen uint64
+	if err := binaryRead(f, &hdrLen); err != nil {
+		return nil, 0, 0, err
+	}
+	hdrBytes := make([]byte, hdrLen)
+	if _, err := io.ReadFull(f, hdrBytes); err != nil {
+		return nil, 0, 0, err
+	}
+
+	var hdr struct {
+		Embeddings struct {
+			Dtype  string   `json:"dtype"`
+			Shape  []int    `json:"shape"`
+			Offset []uint64 `json:"data_offsets"`
+		} `json:"embeddings"`
+	}
+	if err := json.Unmarshal(hdrBytes, &hdr); err != nil {
+		return nil, 0, 0, err
+	}
+	if hdr.Embeddings.Dtype != "F32" {
+		return nil, 0, 0, fmt.Errorf("unsupported dtype %s", hdr.Embeddings.Dtype)
+	}
+	if len(hdr.Embeddings.Shape) != 2 {
+		return nil, 0, 0, fmt.Errorf("expected 2D shape, got %v", hdr.Embeddings.Shape)
+	}
+	vocabSize := hdr.Embeddings.Shape[0]
+	dim := hdr.Embeddings.Shape[1]
+	if len(hdr.Embeddings.Offset) != 2 {
+		return nil, 0, 0, fmt.Errorf("bad offsets")
+	}
+	start := hdr.Embeddings.Offset[0]
+	end := hdr.Embeddings.Offset[1]
+	size := end - start
+	if size != uint64(vocabSize*dim*4) {
+		return nil, 0, 0, fmt.Errorf("size mismatch")
+	}
+
+	if _, err := f.Seek(int64(8+hdrLen+start), io.SeekStart); err != nil {
+		return nil, 0, 0, err
+	}
+
+	buf := make([]byte, size)
+	if _, err := io.ReadFull(f, buf); err != nil {
+		return nil, 0, 0, err
+	}
+
+	mat := make([]float32, vocabSize*dim)
+	for i := 0; i < len(mat); i++ {
+		off := i * 4
+		mat[i] = math.Float32frombits(
+			uint32(buf[off]) |
+				uint32(buf[off+1])<<8 |
+				uint32(buf[off+2])<<16 |
+				uint32(buf[off+3])<<24,
+		)
+	}
+	return mat, vocabSize, dim, nil
+}
+
+func binaryRead(r io.Reader, v any) error {
+	switch p := v.(type) {
+	case *uint64:
+		var b [8]byte
+		if _, err := io.ReadFull(r, b[:]); err != nil {
+			return err
+		}
+		*p = uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
+			uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
+	}
+	return nil
+}
@@ -0,0 +1,59 @@
+package brain
+
+import (
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+)
+
+func modelDir() (string, error) {
+	// 1. Explicit env
+	if v := os.Getenv("KBSEARCH_MODEL"); v != "" {
+		return v, nil
+	}
+	// 2. Next to the binary (dev or installed)
+	self, err := os.Executable()
+	if err == nil {
+		if dir, err := filepath.EvalSymlinks(filepath.Dir(self)); err == nil {
+			cand := filepath.Join(dir, "potion-multilingual-128m")
+			if st, err := os.Stat(cand); err == nil && st.IsDir() {
+				return cand, nil
+			}
+		}
+	}
+	// 3. Repo root lib/ (where other scripts expect it)
+	if v := os.Getenv("KB_ROOT"); v != "" {
+		cand := filepath.Join(v, "lib", "potion-multilingual-128m")
+		if st, err := os.Stat(cand); err == nil && st.IsDir() {
+			return cand, nil
+		}
+	}
+	// 4. HF cache (new layout: models--*/snapshots/*)
+	if v := os.Getenv("HF_HOME"); v != "" {
+		base := filepath.Join(v, "hub")
+		if entries, err := os.ReadDir(base); err == nil {
+			for _, e := range entries {
+				if strings.HasPrefix(e.Name(), "models--") {
+					snapDir := filepath.Join(base, e.Name(), "snapshots")
+					if snaps, err := os.ReadDir(snapDir); err == nil {
+						for _, s := range snaps {
+							cand := filepath.Join(snapDir, s.Name())
+							if st, _ := os.Stat(cand); st != nil && st.IsDir() {
+								return cand, nil
+							}
+						}
+					}
+				}
+			}
+		}
+	}
+	// 5. Legacy HF cache (symlinked model dir)
+	if v := os.Getenv("HF_HOME"); v != "" {
+		cand := filepath.Join(v, "potion-multilingual-128m")
+		if st, err := os.Stat(cand); err == nil && st.IsDir() {
+			return cand, nil
+		}
+	}
+	return "", fmt.Errorf("model not found (set KBSEARCH_MODEL or KB_ROOT, or download to HF cache)")
+}
@@ -0,0 +1,72 @@
+package rank
+
+import (
+	"fmt"
+	"strconv"
+	"strings"
+)
+
+const Usage = `usage: bin/brain/search.go "query" [--root facts|info] [--repo REPO] [-n N] [--json]
+       bin/brain/search.go serve [port]
+       bin/brain/search.go --list-model`
+
+type Options struct {
+	Query     string
+	Root      string
+	Repo      string
+	Limit     int
+	JSONOut   bool
+	ListModel bool
+}
+
+// ParseArgs reads flags. Unknown flags are an error: silently dropping them
+// meant `--hop 1` vanished and its argument `1` was appended to the query.
+// --hop is recognised so it cannot be swallowed; it is not implemented until
+// File/FROM_FILE edges exist.
+func ParseArgs(args []string) (Options, error) {
+	opt := Options{Limit: 20}
+	var queryArgs []string
+
+	for i := 0; i < len(args); i++ {
+		arg := args[i]
+		wantsValue := arg == "--root" || arg == "--repo" || arg == "-n" || arg == "--hop"
+		if wantsValue && i+1 >= len(args) {
+			return opt, fmt.Errorf("%s needs a value", arg)
+		}
+		switch arg {
+		case "--root":
+			i++
+			opt.Root = args[i]
+			if opt.Root != "facts" && opt.Root != "info" {
+				return opt, fmt.Errorf("--root must be facts or info, got %q", opt.Root)
+			}
+		case "--repo":
+			i++
+			opt.Repo = args[i]
+		case "-n":
+			i++
+			n, err := strconv.Atoi(args[i])
+			if err != nil || n < 1 {
+				return opt, fmt.Errorf("-n must be a positive integer, got %q", args[i])
+			}
+			opt.Limit = n
+		case "--hop":
+			return opt, fmt.Errorf("--hop is not implemented yet (needs File/FROM_FILE edges)")
+		case "--json":
+			opt.JSONOut = true
+		case "--list-model":
+			opt.ListModel = true
+		default:
+			if strings.HasPrefix(arg, "-") {
+				return opt, fmt.Errorf("unknown flag %q", arg)
+			}
+			queryArgs = append(queryArgs, arg)
+		}
+	}
+
+	opt.Query = strings.TrimSpace(strings.Join(queryArgs, " "))
+	if opt.Query == "" && !opt.ListModel {
+		return opt, fmt.Errorf("no query given")
+	}
+	return opt, nil
+}
@@ -0,0 +1,9 @@
+package rank
+
+// BM25 ranks best-first, so the top hits are the *highest* scores; cosine
+// distance ranks best-first ascending. Both mirror kblib.py.
+const FTSStmt = "CALL QUERY_FTS_INDEX('Leaf', 'id', $q) " +
+	"RETURN node.id, node.text, node.root, node.source, score ORDER BY score DESC LIMIT $n"
+
+const VecStmt = "CALL QUERY_VECTOR_INDEX('Leaf', 'Leaf_vec', $q, $n) " +
+	"RETURN node.id, node.text, node.root, node.source, distance ORDER BY distance LIMIT $n"
@@ -0,0 +1,100 @@
+// Package rank is the cgo-free ranking and CLI parsing for brain search.
+// CI can `go test ./rank` without the native ladybug library.
+package rank
+
+import (
+	"sort"
+	"strings"
+)
+
+// Hit is one search result, mirroring the python script's dict shape.
+type Hit struct {
+	ID      string  `json:"id"`
+	Text    string  `json:"text"`
+	Root    string  `json:"root"`
+	Source  string  `json:"-"`
+	Score   float64 `json:"score"`
+	Snippet string  `json:"snippet,omitempty"`
+}
+
+// rrfK dampens the contribution of low ranks; same constant as kblib.py.
+const rrfK = 60
+
+// RankAndFilter fuses the two hit lists, applies --root/--repo, then cuts to
+// limit. Cutting first dropped every matching leaf ranked below the cut, so
+// `--root facts` came back empty whenever info leafs filled the top N.
+// limit <= 0 keeps everything.
+func RankAndFilter(fts, vec []Hit, root, repo string, limit int) []Hit {
+	out := Hybrid(fts, vec, 0)
+	if root != "" {
+		out = FilterRoot(out, root)
+	}
+	if repo != "" {
+		out = FilterRepo(out, repo)
+	}
+	if limit > 0 && len(out) > limit {
+		out = out[:limit]
+	}
+	return out
+}
+
+// Hybrid merges FTS and vector hits by reciprocal rank fusion.
+// limit <= 0 returns the full fused list.
+func Hybrid(fts, vec []Hit, limit int) []Hit {
+	byID := make(map[string]Hit, len(fts)+len(vec))
+	rrf := make(map[string]float64, len(fts)+len(vec))
+
+	for i, h := range fts {
+		byID[h.ID] = h
+		rrf[h.ID] += 1.0 / (rrfK + float64(i+1))
+	}
+	for i, h := range vec {
+		if existing, ok := byID[h.ID]; !ok {
+			byID[h.ID] = h
+		} else if existing.Score == 0 {
+			existing.Score = h.Score
+			byID[h.ID] = existing
+		}
+		rrf[h.ID] += 1.0 / (rrfK + float64(i+1))
+	}
+
+	ids := make([]string, 0, len(rrf))
+	for id := range rrf {
+		ids = append(ids, id)
+	}
+	sort.Slice(ids, func(i, j int) bool {
+		if rrf[ids[i]] != rrf[ids[j]] {
+			return rrf[ids[i]] > rrf[ids[j]]
+		}
+		return ids[i] < ids[j]
+	})
+	if limit > 0 && len(ids) > limit {
+		ids = ids[:limit]
+	}
+
+	out := make([]Hit, 0, len(ids))
+	for _, id := range ids {
+		out = append(out, byID[id])
+	}
+	return out
+}
+
+func FilterRoot(hits []Hit, root string) []Hit {
+	var out []Hit
+	for _, h := range hits {
+		if h.Root == root {
+			out = append(out, h)
+		}
+	}
+	return out
+}
+
+func FilterRepo(hits []Hit, repo string) []Hit {
+	var out []Hit
+	for _, h := range hits {
+		if strings.Contains(h.Source, repo) {
+			out = append(out, h)
+		}
+	}
+	return out
+}
@@ -0,0 +1,149 @@
+// Unit tests for ranking/filtering and CLI parsing (no db, no model, offline).
+package rank
+
+import (
+	"strings"
+	"testing"
+)
+
+func h(id, root, source string) Hit {
+	return Hit{ID: id, Text: id, Root: root, Source: source}
+}
+
+func ids(hits []Hit) []string {
+	out := make([]string, len(hits))
+	for i, hit := range hits {
+		out[i] = hit.ID
+	}
+	return out
+}
+
+func eq(t *testing.T, got []Hit, want ...string) {
+	t.Helper()
+	g := ids(got)
+	if len(g) != len(want) {
+		t.Fatalf("got %v, want %v", g, want)
+	}
+	for i := range want {
+		if g[i] != want[i] {
+			t.Fatalf("got %v, want %v", g, want)
+		}
+	}
+}
+
+// A facts leaf that ranks below the limit in the unfiltered list must still
+// be returned for --root facts. Filtering after truncation loses it.
+func TestRankAndFilterFiltersBeforeLimit(t *testing.T) {
+	fts := []Hit{
+		h("i1", "info", "docs/a.md"),
+		h("i2", "info", "docs/b.md"),
+		h("i3", "info", "docs/c.md"),
+		h("f1", "facts", "docker ps x compose"),
+	}
+	eq(t, RankAndFilter(fts, nil, "facts", "", 2), "f1")
+}
+
+func TestRankAndFilterRepoFiltersBeforeLimit(t *testing.T) {
+	fts := []Hit{
+		h("a", "info", "eSlider/2dph:README.md"),
+		h("b", "info", "eSlider/2dph:PLAN.md"),
+		h("c", "info", "eSlider/ops:compose.yaml"),
+	}
+	eq(t, RankAndFilter(fts, nil, "", "ops", 2), "c")
+}
+
+func TestRankAndFilterTruncatesToLimit(t *testing.T) {
+	fts := []Hit{h("a", "info", "x"), h("b", "info", "x"), h("c", "info", "x")}
+	eq(t, RankAndFilter(fts, nil, "", "", 2), "a", "b")
+}
+
+func TestRankAndFilterLimitZeroKeepsAll(t *testing.T) {
+	fts := []Hit{h("a", "info", "x"), h("b", "info", "x")}
+	eq(t, RankAndFilter(fts, nil, "", "", 0), "a", "b")
+}
+
+func TestHybridFusesBothRetrievers(t *testing.T) {
+	fts := []Hit{h("only-fts", "info", "x"), h("both", "info", "x")}
+	vec := []Hit{h("only-vec", "info", "x"), h("both", "info", "x")}
+	eq(t, Hybrid(fts, vec, 0), "both", "only-fts", "only-vec")
+}
+
+func TestHybridTiesAreDeterministic(t *testing.T) {
+	fts := []Hit{h("b", "info", "x"), h("a", "info", "x")}
+	first := ids(Hybrid(fts, nil, 0))
+	for i := 0; i < 50; i++ {
+		got := ids(Hybrid(fts, nil, 0))
+		for j := range first {
+			if got[j] != first[j] {
+				t.Fatalf("unstable order: %v then %v", first, got)
+			}
+		}
+	}
+}
+
+func TestHybridKeepsVectorScoreForSharedHit(t *testing.T) {
+	fts := []Hit{{ID: "x", Root: "info", Score: 0}}
+	vec := []Hit{{ID: "x", Root: "info", Score: 0.87}}
+	got := Hybrid(fts, vec, 0)
+	if len(got) != 1 || got[0].Score != 0.87 {
+		t.Fatalf("got %+v, want score 0.87", got)
+	}
+}
+
+// The old parser dropped unknown flags and appended their arguments to the
+// query, so `search "q" --hop 1` searched for "q 1". --hop is not implemented
+// here (needs File edges); it must still fail closed instead of changing q.
+func TestParseHopIsNotSwallowedIntoTheQuery(t *testing.T) {
+	_, err := ParseArgs([]string{"what runs on arc-2", "--hop", "1"})
+	if err == nil {
+		t.Fatal("expected --hop to error (not implemented), not be swallowed")
+	}
+	if !strings.Contains(err.Error(), "--hop") {
+		t.Fatalf("error should name --hop, got %v", err)
+	}
+}
+
+func TestParseRejectsUnknownFlags(t *testing.T) {
+	if _, err := ParseArgs([]string{"query", "--nope"}); err == nil {
+		t.Fatal("unknown flag accepted")
+	}
+}
+
+func TestParseRejectsBadValues(t *testing.T) {
+	for _, args := range [][]string{
+		{"q", "-n", "zero"},
+		{"q", "-n", "0"},
+		{"q", "--root", "nonsense"},
+		{"q", "--hop"},
+		{"--json"},
+	} {
+		if _, err := ParseArgs(args); err == nil {
+			t.Errorf("accepted %v", args)
+		}
+	}
+}
+
+func TestParseDefaults(t *testing.T) {
+	opt, err := ParseArgs([]string{"two", "words", "--json"})
+	if err != nil || opt.Query != "two words" || opt.Limit != 20 || !opt.JSONOut {
+		t.Fatalf("got %+v err=%v", opt, err)
+	}
+}
+
+func TestListModelNeedsNoQuery(t *testing.T) {
+	if _, err := ParseArgs([]string{"--list-model"}); err != nil {
+		t.Fatalf("unexpected error: %v", err)
+	}
+}
+
+func TestUsageNamesBrainSearch(t *testing.T) {
+	if !strings.Contains(Usage, "bin/brain/search.go") {
+		t.Fatalf("usage must name bin/brain/search.go, got:\n%s", Usage)
+	}
+}
+
+func TestFTSQueryOrdersByScoreDescending(t *testing.T) {
+	if !strings.Contains(FTSStmt, "ORDER BY score DESC") {
+		t.Fatalf("FTS query must order by score DESC, got:\n%s", FTSStmt)
+	}
+}
@@ -0,0 +1,382 @@
+//go:build cgo && system_ladybug
+
+package brain
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"log"
+	"net"
+	"net/http"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strconv"
+	"syscall"
+	"time"
+
+	lbug "github.com/LadybugDB/go-ladybug"
+	"github.com/eSlider/2dph/internal/brain/rank"
+)
+
+const defaultPort = 17830
+const daemonPath = "/embed"
+const healthPath = "/health"
+
+func runSearch(args []string) int {
+	opt, err := rank.ParseArgs(args)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "brain/search: %v\n%s\n", err, rank.Usage)
+		return 2
+	}
+	root, repo, limit, query := opt.Root, opt.Repo, opt.Limit, opt.Query
+	jsonOut := opt.JSONOut
+
+	if opt.ListModel {
+		dir, err := modelDir()
+		if err != nil {
+			fmt.Fprintln(os.Stderr, err)
+			return 1
+		}
+		fmt.Println(dir)
+		return 0
+	}
+
+	if err := openBrain(); err != nil {
+		fmt.Fprintf(os.Stderr, "open brain: %v\n", err)
+		return 1
+	}
+	defer closeBrain()
+
+	hits, err := searchHits(query, root, repo, limit)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "search: %v\n", err)
+		return 1
+	}
+
+	results := hits
+	for i := range results {
+		if results[i].Text != "" {
+			runes := []rune(results[i].Text)
+			if len(runes) > 280 {
+				runes = runes[:280]
+			}
+			results[i].Snippet = string(runes)
+		}
+	}
+
+	out := Dict{
+		{"query", query},
+		{"root_filter", root},
+		{"count", len(results)},
+		{"results", resultsToDicts(results)},
+	}
+
+	if jsonOut {
+		enc := json.NewEncoder(os.Stdout)
+		enc.SetIndent("", "  ")
+		enc.SetEscapeHTML(false)
+		return b2i(enc.Encode(toJSONOut(results, query, root)))
+	}
+	fmt.Print(toYAML(out, 0))
+	return 0
+}
+
+func searchHits(query, root, repo string, limit int) ([]Hit, error) {
+	emb, err := embedQuery(query)
+	if err != nil {
+		return nil, fmt.Errorf("embed: %w", err)
+	}
+	fts, err := queryFTS(query, limit*3)
+	if err != nil {
+		return nil, fmt.Errorf("fts: %w", err)
+	}
+	var vec []Hit
+	if vec, err = queryVector(emb, limit*3); err != nil {
+		fmt.Fprintf(os.Stderr, "vec: %v\n", err)
+	}
+	return rank.RankAndFilter(fts, vec, root, repo, limit), nil
+}
+
+func b2i(err error) int {
+	if err != nil {
+		return 1
+	}
+	return 0
+}
+
+func queryFTS(text string, limit int) ([]Hit, error) {
+	stmt, err := conn.Prepare(rank.FTSStmt)
+	if err != nil {
+		return nil, err
+	}
+	defer stmt.Close()
+	res, err := conn.Execute(stmt, map[string]any{"q": text, "n": limit})
+	if err != nil {
+		return nil, err
+	}
+	return rowsToHits(res)
+}
+
+func queryVector(emb []float64, limit int) ([]Hit, error) {
+	embList := make([]any, len(emb))
+	for i, v := range emb {
+		embList[i] = v
+	}
+	stmt, err := conn.Prepare(rank.VecStmt)
+	if err != nil {
+		return nil, err
+	}
+	defer stmt.Close()
+	res, err := conn.Execute(stmt, map[string]any{"q": embList, "n": limit})
+	if err != nil {
+		return nil, err
+	}
+	hits, err := rowsToHits(res)
+	if err != nil {
+		return nil, err
+	}
+	for i := range hits {
+		hits[i].Score = 1.0 - hits[i].Score
+	}
+	return hits, nil
+}
+
+func rowsToHits(res *lbug.QueryResult) ([]Hit, error) {
+	var hits []Hit
+	for res.HasNext() {
+		row, err := res.Next()
+		if err != nil {
+			return nil, err
+		}
+		vals, err := row.GetAsSlice()
+		if err != nil || len(vals) < 5 {
+			continue
+		}
+		id := fmt.Sprint(vals[0])
+		text := fmt.Sprint(vals[1])
+		root := fmt.Sprint(vals[2])
+		source := fmt.Sprint(vals[3])
+		score := float64(vals[4].(float64))
+		hits = append(hits, Hit{ID: id, Text: text, Root: root, Source: source, Score: score})
+	}
+	return hits, nil
+}
+
+// JSON output types
+type jsonOut struct {
+	Query      string    `json:"query"`
+	RootFilter string    `json:"root_filter"`
+	Count      int       `json:"count"`
+	Results    []jsonHit `json:"results"`
+}
+
+type jsonHit struct {
+	ID      string  `json:"id"`
+	Text    string  `json:"text"`
+	Root    string  `json:"root"`
+	Score   float64 `json:"score"`
+	Snippet string  `json:"snippet,omitempty"`
+}
+
+func toJSONOut(hits []Hit, query, rootFilter string) *jsonOut {
+	out := make([]jsonHit, len(hits))
+	for i, h := range hits {
+		out[i] = jsonHit{
+			ID:      h.ID,
+			Text:    h.Text,
+			Root:    h.Root,
+			Score:   h.Score,
+			Snippet: h.Snippet,
+		}
+	}
+	return &jsonOut{
+		Query:      query,
+		RootFilter: rootFilter,
+		Count:      len(hits),
+		Results:    out,
+	}
+}
+
+func resultsToDicts(hits []Hit) []any {
+	out := make([]any, len(hits))
+	for i, h := range hits {
+		d := Dict{
+			{"id", h.ID},
+			{"text", h.Text},
+			{"root", h.Root},
+			{"score", h.Score},
+		}
+		if h.Snippet != "" {
+			d = append(d, KV{"snippet", h.Snippet})
+		}
+		out[i] = d
+	}
+	return out
+}
+
+// --- Daemon server ---
+func serve(port int) error {
+	model, err := loadModel()
+	if err != nil {
+		return fmt.Errorf("load model: %w", err)
+	}
+	defer model.Close()
+
+	mux := http.NewServeMux()
+	mux.HandleFunc(healthPath, func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusOK)
+	})
+	mux.HandleFunc(daemonPath, func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodPost {
+			w.WriteHeader(http.StatusMethodNotAllowed)
+			return
+		}
+		var req struct {
+			Text string `json:"text"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+			w.WriteHeader(http.StatusBadRequest)
+			return
+		}
+		vec, err := model.Embed(req.Text)
+		if err != nil {
+			w.WriteHeader(http.StatusInternalServerError)
+			json.NewEncoder(w).Encode(map[string]string{"error": err.Error()})
+			return
+		}
+		json.NewEncoder(w).Encode(map[string]any{"vector": vec})
+	})
+
+	addr := fmt.Sprintf("127.0.0.1:%d", port)
+	log.Printf("brain search daemon listening on %s", addr)
+	return http.ListenAndServe(addr, mux)
+}
+
+// --- Daemon client ---
+var daemonClient = &http.Client{
+	Timeout: 5 * time.Second,
+	Transport: &http.Transport{
+		DialContext: (&net.Dialer{Timeout: 2 * time.Second}).DialContext,
+	},
+}
+
+func embedQuery(text string) ([]float64, error) {
+	port := defaultPort
+	if envPort := os.Getenv("KBSEARCH_PORT"); envPort != "" {
+		if p, err := strconv.Atoi(envPort); err == nil {
+			port = p
+		}
+	}
+	if emb, err := tryDaemon(text, port); err == nil {
+		return emb, nil
+	}
+	if os.Getenv("KBSEARCH_NO_DAEMON") == "" {
+		if err := ensureDaemon(port); err == nil {
+			if emb, err := tryDaemon(text, port); err == nil {
+				return emb, nil
+			}
+		}
+	}
+
+	model, err := loadModel()
+	if err != nil {
+		return nil, fmt.Errorf("fallback load model: %w", err)
+	}
+	defer model.Close()
+	return model.Embed(text)
+}
+
+func tryDaemon(text string, port int) ([]float64, error) {
+	url := fmt.Sprintf("http://127.0.0.1:%d%s", port, daemonPath)
+	payload := map[string]string{"text": text}
+	body, _ := json.Marshal(payload)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	req, _ := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := daemonClient.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("daemon HTTP %d", resp.StatusCode)
+	}
+	var r struct {
+		Vector []float64 `json:"vector"`
+		Error  string    `json:"error"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&r); err != nil {
+		return nil, err
+	}
+	if r.Error != "" {
+		return nil, errors.New(r.Error)
+	}
+	return r.Vector, nil
+}
+
+func ensureDaemon(port int) error {
+	url := fmt.Sprintf("http://127.0.0.1:%d%s", port, healthPath)
+	ctx, cancel := context.WithTimeout(context.Background(), 500*time.Millisecond)
+	defer cancel()
+	req, _ := http.NewRequestWithContext(ctx, "GET", url, nil)
+	if resp, err := daemonClient.Do(req); err == nil {
+		resp.Body.Close()
+		if resp.StatusCode == http.StatusOK {
+			return nil
+		}
+	}
+
+	self, err := os.Executable()
+	if err != nil {
+		return err
+	}
+	cmd := exec.Command(self, "serve", strconv.Itoa(port))
+	cmd.Dir, _ = filepath.Split(self)
+	cmd.Stdout = nil
+	cmd.Stderr = nil
+	cmd.SysProcAttr = &syscall.SysProcAttr{Setsid: true}
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+	if cmd.Process != nil {
+		_ = cmd.Process.Release()
+	}
+
+	for i := 0; i < 40; i++ {
+		time.Sleep(250 * time.Millisecond)
+		req, _ := http.NewRequestWithContext(context.Background(), "GET", url, nil)
+		if resp, err := daemonClient.Do(req); err == nil {
+			resp.Body.Close()
+			if resp.StatusCode == http.StatusOK {
+				return nil
+			}
+		}
+	}
+	return fmt.Errorf("daemon failed to start on port %d", port)
+}
+
+// Main is the bin/brain/search.go entry: search, serve, or --list-model.
+func Main(args []string) int {
+	if len(args) > 0 && args[0] == "serve" {
+		port := defaultPort
+		if len(args) > 1 {
+			if p, err := strconv.Atoi(args[1]); err == nil {
+				port = p
+			}
+		}
+		if err := serve(port); err != nil {
+			log.Printf("brain/search serve: %v", err)
+			return 1
+		}
+		return 0
+	}
+	return runSearch(args)
+}
@@ -0,0 +1,13 @@
+package brain
+
+import (
+	"os"
+
+	"github.com/eSlider/2dph/internal/brain/rank"
+)
+
+func eps() string { return os.Getenv("KBTEST_EPS") }
+
+// Hit is the search hit type; ranking lives in package rank so CI can test
+// it without the native ladybug library.
+type Hit = rank.Hit
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
eSlider	c9283ce0a9	feat: in-process HTTP search; /get /stats /audit /ingest. Tests / Test (push) Skipped Tests / Release (semver) (push) Skipped bin/brain/serve.go (ladybug tags) calls internal/brain instead of exec. HTTP tests inject a fake API so CI stays cgo-free. ExecSearcher remains the fallback when the binary is built without system_ladybug.	2026-08-13 17:50:46 +01:00
eSliderandGitHub	20b78a9a20	feat: brain/index.go shebang; mail import is not a brain write (D14). (#12 ) Tests / Test (push) Failing after 5s Details Tests / Release (semver) (push) Skipped Details Commands live at bin/brain/{index,get,stats,eval,watch}.go and bin/mail/import.go, bin/markdown/import.go, bin/postgres/query.go. Python remains the Ladybug write worker. index_mail is a deprecation shim that rebuilds via --with-mail.	2026-08-13 17:46:25 +01:00
eSliderandGitHub	5d4b3427a4	refactor: chats method shebangs; drop chats index (D14). (#11 ) Tests / Test (push) Failing after 5s Details Tests / Release (semver) (push) Skipped Details Parsers and commands live in internal/chats. bin/chats/{sync,import,facts,apply}.go are tagged shebang mains. Brain ingest is not a chats command.	2026-08-13 17:31:03 +01:00
eSliderandGitHub	1c7db6d499	docs: name bin/brain/search.go; --hop is not a graph walk. (#10 ) Tests / Test (push) Failing after 6s Details Tests / Release (semver) (push) Skipped Details Published docs and skills still taught bin/kb/search --hop 1. Search lives at bin/brain/search.go; --hop errors until File edges exist. A unittest gates the SoT so the lie cannot return.	2026-08-13 17:23:40 +01:00
eSliderandGitHub	0786ddcb06	feat: bin/brain/serve.go; search backend is Go not Python (#9 ) Tests / Test (push) Failing after 5s Details Tests / Release (semver) (push) Skipped Details * feat(brain): HTTP serve from bin/brain/serve.go, default Go search binary. Move the HTTP package to internal/httpapi. Default backend is var/bin/brain-search, not Python. bin/serve.go stays as a deprecation shim. * feat(httpapi): default search backend is var/bin/brain-search. bin/brain/serve.go is the command; bin/serve.go stays as a tagged deprecation shim. Tests fail if the default path still names Python.	2026-08-13 14:32:21 +01:00
eSliderandGitHub	eeb5b79cf2	refactor: one Go module; brain search in bin/brain + internal/brain. (#8 ) Tests / Test (push) Failing after 5s Details Tests / Release (semver) (push) Skipped Details Collapse nested kbsearch/chats go.mod into the root module. Ranking stays cgo-free under internal/brain/rank so CI does not need ladybug. bin/kb/search is a deprecation wrapper that still sets CGO and builds the binary.	2026-08-13 14:26:54 +01:00
eSliderandGitHub	5990feb1f6	docs: point issues at Gitea origin (D15). (#7 ) Tests / Test (push) Failing after 4s Details Tests / Release (semver) (push) Skipped Details GitHub stays the public clone for PRs and Actions. Work board is https://git.produktor.io/eSlider/2dph/issues.	2026-08-13 14:11:00 +01:00
eSliderandGitHub	d27a738fee	feat(chats): parse LinkedIn MCP v4.22 inbox/conversation blobs. (#6 ) Tests / Test (push) Failing after 29s Details Tests / Release (semver) (push) Skipped Details get_inbox/get_conversation return a sections+references envelope, not a message list. Parser is covered by synthetic Alice/Bob fixtures; CI now runs the nested bin/chats tests. Session check no longer launches Chromium.	2026-08-13 12:25:51 +01:00
eSliderandGitHub	669e184cf6	fix(kbsearch): rank FTS correctly, filter before -n, start the daemon. (#5 ) Go search took worst BM25 hits (ORDER BY score), cut to -n before --root, and never called ensureDaemon. Ranking and flag parsing move to a cgo-free package so CI can fail those regressions without ladybug. --hop errors instead of being swallowed into the query.	2026-08-13 12:19:26 +01:00
eSliderandGitHub	ebc3f948c1	Add Gmail --query to mail/sync (default in:inbox) (#4 ) * Add --query to Gmail mail/sync instead of always listing in:inbox. Callers keep the search string; default remains in:inbox. * Document Gmail --query on the mail/sync pipeline. * test(mail): assert Gmail --query reaches ListIDs, not only the CLI flag. ParseCLI coverage left a hole: an empty query still has to become in:inbox and a custom q has to be the string the client lists with.	2026-08-13 12:19:22 +01:00
eSlider	fe6a02024c	feat(chats): LinkedIn source — MCP client via get_inbox + get_conversation - LinkedInMCPSource: MCP JSON-RPC, как TelegramMCPSource - sync linkedin --limit N: выгрузка сообщений из LinkedIn - Проверка сессии: uvx mcp-server-linkedin --status - Вывод инструкции если сессия истекла - JSONL в var/chats/linkedin/<thread_id>/messages.jsonl	2026-08-13 00:10:42 +01:00
eSlider	4a065d9838	docs: add edelweiss to GitHub safety rules	2026-08-13 00:07:01 +01:00
eSlider	e3c6ef5684	chore: remove edelweiss references from public repo	2026-08-13 00:06:49 +01:00
eSlider	98c14e23f1	docs: GitHub safety rules — no absolute paths, PII, secrets, curasoft	2026-08-13 00:02:09 +01:00
eSlider	ff1716de40	fix: resolve plan.md conflict, remove remaining /mnt/ paths	2026-08-13 00:00:48 +01:00
eSlider	fec5325c7a	chore: clean absolute paths, curasoft refs, secrets from history - bin/chats/: env-based paths, no /mnt/ /home/ hardcodes - bin/edelweiss-pilot: remove curasoft, use DOCS_BASE env var - bin/facts/crm: use KNOWLEDGE_MESH_SEED env var - compose.edelweiss.yml: remove curasoft volumes, use DOCS_BASE - docs/chat-import-plan.md: link to Gitea issue, no secrets - bin/seed-edelweiss-facts.py: removed (curasoft-only)	2026-08-13 00:00:15 +01:00
eSlider	27d9521e7f	bin/chats: Phase 1 MVP — Telegram sync/import/index/facts/apply - bin/chats/ — nested Go module (как bin/kbsearch/) - sync telegram — MCP JSON-RPC клиент, 31 личный чат, 922 сообщения - import — конвертация JSONL → MD с YAML frontmatter - index — делегирует bin/kb/index --corpus (132 leafs в brain) - facts — regex extraction phone/email/linkedin с валидацией (исключены: даты, суммы, номера карт, инвойсы) - apply — oo CLI cross-check + dry-run - Source interface для будущих WhatsApp/LinkedIn - 4 system tests (import, facts, empty, roundtrip) — синтетические данные - bin/chat — build+exec wrapper - docs/chat-import-plan.md — прогресс, пути к env (без секретов) Безопасность: var/ в gitignore, credentials в env, тесты без реальных данных.	2026-08-12 23:59:29 +01:00
eSlider	a7cb8d4c76	docs: chat import pipeline plan — link to Gitea issue #1	2026-08-12 23:59:20 +01:00
eSliderandCursor	53cd00284d	fix(kb): seed facts before CREATE indexes (FTS MERGE corruption) Upsert under live FTS raises "document for node offset N is missing". Add --skip-indexes; edelweiss-pilot index = write → seed → ensure_indexes. Ship seed-edelweiss-facts.py (paired lexicon/OO/interview/QEMU facts). Co-authored-by: Cursor <cursoragent@cursor.com>	2026-08-12 16:09:02 +01:00
eSliderandCursor	b73b4d4f97	fix(kb): stop DROP INDEX killing HNSW via Ladybug ghost catalog Ladybug 0.19 DROP INDEX leaves `_0_Leaf_vec_UPPER` / `0_id_docs` in catalog so CREATE fails while SHOW_INDEXES omits the index; create_fts_and_vector used to swallow that. Never drop FTS/VECTOR; ensure_indexes after upserts; rebuild = delete kb.lbug. Add compose.edelweiss.yml + regression tests. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-08-12 16:05:14 +01:00
eSlider	1d1f6a90ff	Remove curasoft references, rename to detective method - PLAN.md: replace 'curasoft-detective' with 'detective method' - README.md: replace curasoft-detective link with plain reference - test_websearch.py: fix test domain from ticket.curasoft.de to example.com - Rewrote git history with git-filter-repo to remove all traces	2026-08-12 13:46:54 +01:00
eSlider	f220bcd95a	kbsearch: Go implementation with daemon model serving - New nested module bin/kbsearch with Go implementation of bin/kb/search - Embedding model (potion-multilingual-128M) served by localhost daemon so repeated CLI calls reuse the loaded model - Bash launcher bin/kb/search builds binary on first run, caches to var/bin/ - Hybrid FTS + vector search (RRF k=60) matching Python kblib behavior - YAML output via port of yamlout.py (ordered keys, same format) - JSON output with proper field order - All flags: --root, --repo, -n, --json, --list-model - Root go.mod reverted to 1.25.0 (kbsearch is isolated nested module) - CI passes: go test ./... and go vet ./... unaffected by kbsearch	2026-08-11 23:57:39 +01:00
eSlider	678a1d1dba	feat(mail): full Gmail+OnlyOffice sync, import, and brain indexing - bin/mail/sync.go: async Go sync engine (8 workers, paginated Gmail via API + OnlyOffice IMAP); Gmail attachments key off body.attachmentId, not MIME partId; ICS sidecars Latin-1->UTF-8 normalized (TestICSToMarkdownNormalizesLatin1) - bin/mail/import: message.json -> markdown; PDFs via pdftotext -layout fast path with docling subprocess fallback for the ~5% textless files - bin/mail/index_mail: fresh-rebuild indexer (repo corpus + mail) avoiding ladybug WAL corruption on bulk-insert into indexed DBs; split from import - bin/kb/index: keep FTS/VECTOR indexes across incremental runs (drop+recreate leaves stale backing tables killing the vector index) - docs: README/PLAN/AGENTS cover the mail pipeline Result: 17,835 messages -> 28,918 info leafs, FTS+HNSW healthy.	2026-08-11 21:57:38 +01:00
eSlider	8781c0c3eb	refactor(tools): bin/{subject}/{method} layout; Go serve+watch modules Move serve/ (module) -> bin/server, tools/ -> bin/tools, replace bin/kb-watch bash with bin/watch Go package; self-executing Go shebangs bin/serve.go and bin/kb/watch.go; Docker + CI + git/import + docs repointed. Multi-stage image builds static serve+watch binaries (no Go runtime in container).	2026-08-11 09:52:20 +01:00
eSlider	d6b17e8819	feat(kb): CRM association proof via oo, fix ssh-tunnel self-ref + oo creds - bin/facts/crm: prove person<->company/company<->project against ooCRM x corpus SoT (knowledge-mesh-seed.yaml), write 78 facts (root=facts) - tools/crmfacts.py + test_crm_facts.py: parser under unit tests (26 pass) - docs/crm-associations-proof.md: provable graph, mistakes, fixes - oo merge 759->763 resolves duplicate GoldenRatio.Exchange legal entity - bin/db/ssh-tunnel: "$0" self-check + accept-new/BatchMode ssh flags - AGENTS.md: document bin/facts/crm	2026-08-10 23:22:34 +01:00