docs: name bin/brain/search.go; --hop is not a graph walk. (#10)

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.

.github/workflows/ci.yml

@@ -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:
@@ -32,16 +36,20 @@ jobs:
           bash -n bin/db/psql-yq
           bash -n bin/db/ssh-tunnel
           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 bin/tools -t .
 
-      - name: Go tests (server + watch packages)
+      - name: Go tests (root module, no ladybug cgo)
         run: |
           go vet ./...
           go test ./... -count=1
 
+      - 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: |
           ./bin/facts/audit self 2>/dev/null || echo "audit: not yet implemented; gate skipped"

.gitignore

@@ -10,3 +10,4 @@ __pycache__/
 .env
 .secrets/
 lib-ladybug/
+go.work.local

AGENTS.md

@@ -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.
@@ -35,10 +35,10 @@ 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/          self-describing tools bin/{subject}/{method}.go (shebang)
-bin/serve.go  async Go HTTP server entry (self-executing go run shebang)
+bin/brain/    search.go, serve.go; libs in internal/brain and internal/httpapi
-bin/watch/    corpus watcher Go package (mtimes, no inotify deps)
+internal/     shared Go (brain/rank is cgo-free)
-bin/server/   async Go HTTP server (goroutines, bounded worker pool)
+bin/watch/    corpus watcher (internal via bin/brain/watch later)
 bin/mail/     mail pipeline: sync (Go), import (md), index_mail (rebuild)
 bin/tools/    vendored python libs behind bin/* (kblib, yamlout, websearch)
 bin/docker-entrypoint  container entrypoint (brain index|search|serve|watch)
@@ -52,6 +52,7 @@ var/          kb.lbug, var/mail/*, caches (gitignored)
 
 ```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 --from-raw var/mail                                     # message.json → message.md (convert only)
 bin/mail/index_mail                                                     # rebuild brain incl. all mail (fresh DB)
 ```
@@ -73,7 +74,8 @@ bin/mail/index_mail                                                     # rebuil
 ```bash
 bin/facts/audit ["self"|"facts"|"info"|"stale"]   # 2-source + staleness gate
 bin/facts/crm [--dry-run]                         # proof person↔company/company↔project (ooCRM × corpus SoT)
-bin/kb/search "query" [--hop N] [--repo X]        # deduction search → YAML
+bin/kb/search "query" [--repo X]                  # deprecated wrapper → bin/brain/search.go
+bin/brain/search.go "query" [--root facts|info]   # deduction search → YAML
 bin/md/tables                                     # what the graph holds → YAML
 bin/brain/deduce "question"                       # thinking wrapper
 ```
@@ -89,7 +91,7 @@ parameter. Search before reading whole files.
    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 — no files, no mentions.** Remove all traces if found.
+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.

PLAN.md

@@ -26,10 +26,10 @@ detective method: **a fact needs ≥2 independent sources or it is
 |---|----------|--------|
 | 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-dependent skills. |
-| D3 | web search | import `web-search`, retire local `searxng-ops`. Vendored here, no remote link. |
+| 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`, `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`. |
+| 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 | GitHub `eSlider/2dph`, public (like sibling repos), push/commit via `gh`, TDD + commit every change, CI/CD. |
+| 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
 
@@ -51,9 +53,10 @@ 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/index                build FTS + HNSW from corpus (Python, for now)
-    kb/search               deduction: facts → info → web-search; --hop N
+    brain/search.go         deduction: facts → info → web-search
     kb/get  kb/stats  kb/eval
+    brain/serve.go          HTTP API (internal/httpapi)
     md/import  md/select  md/tables  md/gaps     (mistune)
     brain/extract  brain/audit   brain/deduce    (thinking wrapper)
     web/search               (vendored)
@@ -110,17 +113,18 @@ Common props on every node/edge: `root`, `confidence`, `evidence[]`, `how`,
    corrupts its WAL on bulk-insert into an already-indexed DB. Conversion and
    indexing stay separate for crash safety.
 4. Result: 17,835 messages → 28,918 info leafs, FTS + HNSW healthy, searchable
-   via `bin/kb/search`.
+   via `bin/brain/search.go`.
 
 ## CI/CD pipeline (D15)
 
 `.github/workflows/ci.yml`:
 
-1. go vet + go test ./... (Go tools)
+1. go vet + go test ./... (root module; packages without ladybug cgo)
-2. python -m unittest discover + pytest (Py tools)
+2. `go test ./internal/brain/rank` (cgo-free ranking + flag parser)
-3. bin/facts/audit self  (lexicon internal consistency)
+3. python -m unittest discover -s bin/tools (includes published-docs SoT)
-4. bin/kb/eval            (recall@5 ≥ 0.95, gates index regressions)
+4. bin/facts/audit self  (lexicon internal consistency)
-5. md-docs build/lint if docs tooling arrives.
+5. bin/kb/eval            (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.

README.md

@@ -32,7 +32,7 @@ graph TB
         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"]
+        SR["bin/brain/search.go<br/>deduction"]
     end
 
     subgraph store["Ladybug var/kb.lbug"]
@@ -85,21 +85,23 @@ 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/brain/search.go "Matrix federation over HTTPS"   # facts → info → web-search
-bin/kb/search "what runs on arc-2" --hop 1        # walk graph edges
+bin/brain/search.go "onlyoffice postgres" --root facts
-bin/kb/search "where is cs-lexicon" --json | yq '.'  # YAML by default
+bin/brain/search.go "where is cs-lexicon" --json | yq '.'
-bin/kb/get <id> --body                            # full chunk on demand
+bin/kb/get <id> --body                               # full chunk on demand
-bin/kb/stats                                      # index health
+bin/kb/stats                                         # index health
-bin/kb/eval                                       # recall@5 gate
+bin/kb/eval                                          # 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 --from-raw var/mail                                     # JSON → markdown
 bin/mail/index_mail                                                     # rebuild brain incl. mail
-bin/kb/search "Mietwagen Nürnberg invoice"                              # now answers from mail
+bin/brain/search.go "invoice from last week"                            # same search over mail leafs
 ```
 
 ## Storage
@@ -117,10 +119,10 @@ bin/kb/search "Mietwagen Nürnberg invoice"                              # now a
 
 ## Tooling conventions
 
-`bin/{subject}/{method}` — self-describing: shebang on line 1, usage comment
+`bin/{subject}/{method}.go` — self-describing: shebang on line 1, usage comment
-from line 2. bash + python primary; golang via the Go shebang when a compiled
+from line 2. Shared code in `internal/`. YAML default output, `--json` for
-helper is right. YAML default output, `--json` for machines. Everything that
+machines. Tests gate every commit. HTTP: `bin/brain/serve.go` (default search
-touches network/db is read-only, throttled, cached. Tests gate every commit.
+binary `var/bin/brain-search`, not Python).
 
 ## Development
 
@@ -136,7 +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
 ```
 
@@ -148,4 +150,7 @@ docker compose up brain-watch                  # auto re-index on change
   skills (`web-search`, `db-yaml`, …) that 2dph integrates
 - detective method — the two-source method
 
+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.

bin/brain/doc.go

@@ -0,0 +1,3 @@
+// Commands in this directory are shebang mains (search.go).
+// search.go is behind the system_ladybug build tag (cgo).
+package main

bin/brain/search.go

@@ -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:]))
+}

bin/brain/serve.go

@@ -0,0 +1,26 @@
+//usr/bin/env go run -tags=brain_serve "$0" "$@"; exit
+//go:build brain_serve
+//
+// bin/brain/serve.go - HTTP API for the 2dph brain.
+//
+//	KB_ROOT=/path/to/2dph ./bin/brain/serve.go
+//	KB_SEARCH_CMD=... KB_WORKERS=4 KB_PORT=8630 ./bin/brain/serve.go
+//
+// Default search backend is var/bin/brain-search (Go), not Python.
+// NOTE: never run `gofmt -w` on this file — it breaks the shebang.
+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()
+}

bin/chats/go.mod

@@ -1,3 +0,0 @@
-module github.com/eSlider/2dph/bin/chats
-
-go 1.25.0

bin/chats/linkedin.go

@@ -0,0 +1,669 @@
+package main
+
+import (
+	"bufio"
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"regexp"
+	"strings"
+	"time"
+)
+
+type LinkedInMCPSource struct {
+	userDataDir string
+	limit       int
+}
+
+type lnInboxItem struct {
+	ThreadID         string `json:"thread_id"`
+	Participants     string `json:"participants"`
+	LastMessage      string `json:"last_message"`
+	LastMessageDate  string `json:"last_message_date"`
+	Unread           bool   `json:"unread"`
+}
+
+// mcp-server-linkedin v4.22 returns get_inbox / get_conversation as
+// {url, sections:{inbox|conversation: textblob}, references:{...}}.
+// The conversation list lives in references (kind=conversation); messages live
+// in the sections text blob, delimited by "<From> sent the following message
+// at <time>" markers. See testdata/linkedin_*.json for the wire shape.
+
+type lnEnvelope struct {
+	URL        string          `json:"url"`
+	Sections   map[string]any  `json:"sections"`
+	References map[string]any  `json:"references"`
+}
+
+type lnReference struct {
+	Kind    string `json:"kind"`
+	URL     string `json:"url"`
+	Text    string `json:"text"`
+	Context string `json:"context"`
+}
+
+type lnMessage struct {
+	From string `json:"from"`
+	Date string `json:"date"`
+	Text string `json:"text"`
+}
+
+var (
+	lnWeekdays = map[string]time.Weekday{
+		"SUNDAY": time.Sunday, "MONDAY": time.Monday, "TUESDAY": time.Tuesday,
+		"WEDNESDAY": time.Wednesday, "THURSDAY": time.Thursday,
+		"FRIDAY": time.Friday, "SATURDAY": time.Saturday,
+	}
+	lnMsgStartRe = regexp.MustCompile(`^(.+?) sent the following messages? at (.+)$`)
+	lnTimeRe     = regexp.MustCompile(`\d{1,2}:\d{2}\s*[AP]M`)
+)
+
+func isWeekdayLine(s string) bool {
+	if _, ok := lnWeekdays[s]; ok {
+		return true
+	}
+	switch s {
+	case "TODAY", "YESTERDAY", "THIS WEEK", "LAST WEEK":
+		return true
+	}
+	return lnMonthDayRe.MatchString(s)
+}
+
+var lnMonthDayRe = regexp.MustCompile(`^[A-Z]{3}\s+\d{1,2}$`)
+
+// parseLinkedInInbox extracts conversations from a get_inbox response.
+func parseLinkedInInbox(text string) []lnInboxItem {
+	var env lnEnvelope
+	if err := json.Unmarshal([]byte(text), &env); err != nil {
+		return nil
+	}
+	refs, _ := env.References["inbox"].([]any)
+	var items []lnInboxItem
+	for _, r := range refs {
+		rr, ok := r.(map[string]any)
+		if !ok {
+			continue
+		}
+		if rr["kind"] != "conversation" {
+			continue
+		}
+		u, _ := rr["url"].(string)
+		tid := threadIDFromURL(u)
+		if !validThreadID(tid) {
+			continue
+		}
+		name, _ := rr["text"].(string)
+		items = append(items, lnInboxItem{
+			ThreadID:     tid,
+			Participants: name,
+		})
+	}
+	return items
+}
+
+// parseLinkedInConversation parses the sections.conversation text blob into
+// messages. Messages are delimited by "<From> sent the following message(s) at
+// <time>" lines; each message body runs until the next marker. Day headers
+// (all-caps weekdays) provide date context; times are mapped to the most
+// recent matching weekday.
+func parseLinkedInConversation(text string) []lnMessage {
+	var env lnEnvelope
+	if err := json.Unmarshal([]byte(text), &env); err != nil {
+		return nil
+	}
+	blob, _ := env.Sections["conversation"].(string)
+	if blob == "" {
+		return nil
+	}
+
+	var msgs []lnMessage
+	var cur *lnMessage
+	var body []string
+	day := ""
+
+	flush := func() {
+		if cur == nil {
+			return
+		}
+		cur.Text = strings.TrimSpace(strings.Join(body, "\n"))
+		if ts := linkedInTimestamp(day, cur.Date); ts != "" {
+			cur.Date = ts
+		}
+		if cur.Text != "" {
+			msgs = append(msgs, *cur)
+		}
+		cur = nil
+		body = nil
+	}
+
+	for _, raw := range strings.Split(blob, "\n") {
+		line := strings.TrimSpace(raw)
+		if line == "" {
+			continue
+		}
+		if isWeekdayLine(line) {
+			if line != day {
+				// A new day header terminates the previous message,
+				// which must keep the earlier date context.
+				flush()
+			}
+			day = line
+			continue
+		}
+		if m := lnMsgStartRe.FindStringSubmatch(line); m != nil {
+			flush()
+			cur = &lnMessage{From: strings.TrimSpace(m[1]), Date: strings.TrimSpace(m[2])}
+			continue
+		}
+		if cur == nil {
+			continue
+		}
+		// Skip "View X's profile" and the "<From> (pronouns) <time>" header.
+		if strings.HasPrefix(line, "View ") && strings.HasSuffix(line, "'s profile") {
+			continue
+		}
+		if strings.HasPrefix(line, cur.From) && lnTimeRe.MatchString(line) {
+			continue
+		}
+		body = append(body, line)
+	}
+	flush()
+	return msgs
+}
+
+// linkedInTimestamp maps a weekday, relative, or MON DD date header + clock
+// string to a timestamp, or returns "" when the clock cannot be parsed.
+func linkedInTimestamp(day, clock string) string {
+	t, err := time.Parse("3:04 PM", clock)
+	if err != nil {
+		return ""
+	}
+	now := time.Now()
+	var d time.Time
+	if wd, ok := lnWeekdays[day]; ok {
+		diff := (int(now.Weekday()) - int(wd) + 7) % 7
+		d = now.AddDate(0, 0, -diff)
+	} else {
+		switch day {
+		case "TODAY":
+			d = now
+		case "YESTERDAY":
+			d = now.AddDate(0, 0, -1)
+		case "THIS WEEK":
+			diff := int(now.Weekday())
+			d = now.AddDate(0, 0, -diff)
+		case "LAST WEEK":
+			diff := int(now.Weekday()) + 7
+			d = now.AddDate(0, 0, -diff)
+		default:
+			if m := lnMonthDayRe.FindStringSubmatch(day); m != nil {
+				// MON DD without a year: resolve to the most recent
+				// occurrence that is not in the future.
+				d = monthDayDate(day, now)
+				if d.IsZero() {
+					return t.Format("15:04")
+				}
+			} else {
+				// No date context; keep bare clock time.
+				return t.Format("15:04")
+			}
+		}
+	}
+	res := time.Date(d.Year(), d.Month(), d.Day(), t.Hour(), t.Minute(), 0, 0, time.UTC)
+	return res.UTC().Format(time.RFC3339)
+}
+
+var lnMonths = map[string]time.Month{
+	"JAN": time.January, "FEB": time.February, "MAR": time.March,
+	"APR": time.April, "MAY": time.May, "JUN": time.June,
+	"JUL": time.July, "AUG": time.August, "SEP": time.September,
+	"OCT": time.October, "NOV": time.November, "DEC": time.December,
+}
+
+// monthDayDate resolves "MON DD" to the most recent occurrence of that date,
+// preferring the current year and falling back to the previous year when the
+// date is in the future. Returns zero time when unresolvable.
+func monthDayDate(day string, now time.Time) time.Time {
+	parts := strings.Fields(day)
+	if len(parts) != 2 {
+		return time.Time{}
+	}
+	mo, ok := lnMonths[parts[0]]
+	if !ok {
+		return time.Time{}
+	}
+	var dd int
+	if _, err := fmt.Sscanf(parts[1], "%d", &dd); err != nil {
+		return time.Time{}
+	}
+	if dd < 1 || dd > 31 {
+		return time.Time{}
+	}
+	d := time.Date(now.Year(), mo, dd, 0, 0, 0, 0, time.UTC)
+	if d.After(now) {
+		d = d.AddDate(-1, 0, 0)
+	}
+	if d.After(now) {
+		return time.Time{}
+	}
+	return d
+}
+
+func threadIDFromURL(u string) string {
+	u = strings.TrimSuffix(u, "/")
+	idx := strings.LastIndex(u, "/")
+	if idx < 0 {
+		return ""
+	}
+	return u[idx+1:]
+}
+
+// validThreadID rejects path segments that are not real thread ids (e.g. the
+// literal "thread" or an empty trailing segment).
+func validThreadID(id string) bool {
+	if id == "" || id == "thread" {
+		return false
+	}
+	return true
+}
+
+func NewLinkedInMCPSource(userDataDir string) *LinkedInMCPSource {
+	return &LinkedInMCPSource{userDataDir: userDataDir}
+}
+
+func (s *LinkedInMCPSource) Name() string { return "linkedin" }
+
+func (s *LinkedInMCPSource) Sync(ctx context.Context, outDir string, limit int) error {
+	if limit > 0 {
+		s.limit = limit
+	}
+
+	// getConversation fetches one thread, recreating the MCP server when it
+	// wedges. A single 429 makes mcp-server-linkedin close its browser and
+	// refuse every later call ("still has a browser open"), so a broken server
+	// must be restarted rather than hammered.
+	getConversation := func(threadID string) ([]lnMessage, error) {
+		client, err := newLinkedInMCP(ctx, s.userDataDir)
+		if err != nil {
+			return nil, fmt.Errorf("linkedin mcp: %w", err)
+		}
+		defer client.Close()
+		msgs, err := client.GetConversation(ctx, "", threadID, msgLimitFor(s.limit))
+		if err != nil && wedged(err) {
+			fmt.Fprintf(os.Stderr, "chats: %s: server wedged, restarting broker\n", threadID)
+			time.Sleep(5 * time.Second)
+			client2, cerr := newLinkedInMCP(ctx, s.userDataDir)
+			if cerr == nil {
+				defer client2.Close()
+				msgs, err = client2.GetConversation(ctx, "", threadID, msgLimitFor(s.limit))
+			}
+		}
+		return msgs, err
+	}
+
+	client, err := newLinkedInMCP(ctx, s.userDataDir)
+	if err != nil {
+		return fmt.Errorf("linkedin mcp: %w", err)
+	}
+	inbox, err := client.GetInbox(ctx, 50)
+	if err != nil {
+		client.Close()
+		return fmt.Errorf("get_inbox: %w", err)
+	}
+	client.Close()
+	if len(inbox) == 0 {
+		fmt.Println("chats: no LinkedIn conversations found")
+		return nil
+	}
+
+	fmt.Printf("chats: found %d LinkedIn conversations\n", len(inbox))
+
+	for _, conv := range inbox {
+		convID := sanitizeDir(conv.ThreadID)
+		if convID == "" {
+			convID = fmt.Sprintf("conv_%d", time.Now().UnixNano())
+		}
+
+		parts := strings.SplitN(conv.Participants, ",", 2)
+		chatName := strings.TrimSpace(parts[0])
+		if chatName == "" {
+			chatName = convID
+		}
+
+		chatDir := filepath.Join(outDir, "linkedin", convID)
+		if err := os.MkdirAll(chatDir, 0755); err != nil {
+			fmt.Fprintf(os.Stderr, "chats: mkdir %s: %v\n", chatDir, err)
+			continue
+		}
+
+		msgs, err := getConversation(conv.ThreadID)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "chats: get_conversation %s: %v\n", convID, err)
+			continue
+		}
+
+		jsonlPath := filepath.Join(chatDir, "messages.jsonl")
+
+		// A rate-limited response can parse to zero messages. Never clobber
+		// previously synced data with an empty file.
+		if len(msgs) == 0 {
+			fmt.Fprintf(os.Stderr, "chats: %s (%s): 0 messages parsed, keeping existing file\n", chatName, convID)
+			continue
+		}
+
+		f, err := os.Create(jsonlPath)
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "chats: create %s: %v\n", jsonlPath, err)
+			continue
+		}
+
+		enc := json.NewEncoder(f)
+		written := 0
+		for i, m := range msgs {
+			text := m.Text
+			if text == "" {
+				continue
+			}
+
+			ts := m.Date
+			if t, err := time.Parse("2006-01-02T15:04:05Z07:00", m.Date); err == nil {
+				ts = t.UTC().Format(time.RFC3339)
+			} else if t, err := time.Parse(time.RFC3339, m.Date); err == nil {
+				ts = t.UTC().Format(time.RFC3339)
+			}
+
+			chatMsg := Message{
+				ID:        fmt.Sprintf("li_%s_%d", convID, i),
+				Timestamp: ts,
+				From:      m.From,
+				Text:      text,
+				Platform:  "linkedin",
+			}
+			if err := enc.Encode(chatMsg); err != nil {
+				fmt.Fprintf(os.Stderr, "chats: encode: %v\n", err)
+				continue
+			}
+			written++
+		}
+		f.Close()
+
+		fmt.Printf("chats: synced %s (%s) — %d messages\n", chatName, convID, written)
+
+		// Pause between conversations to reduce LinkedIn rate limiting.
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		case <-time.After(2 * time.Second):
+		}
+	}
+
+	return nil
+}
+
+type linkedInMCPClient struct {
+	cmd    *exec.Cmd
+	stdin  *bufio.Writer
+	stdout *bufio.Scanner
+	msgID  int
+}
+
+func newLinkedInMCP(ctx context.Context, userDataDir string) (*linkedInMCPClient, error) {
+	args := []string{
+		"mcp-server-linkedin@latest",
+		"--user-data-dir", userDataDir,
+		"--no-auto-import",
+		"--no-daemon",
+		"--transport", "stdio",
+		"--login-timeout", "10",
+		"--browser-wait", "1",
+		"--browser-idle-timeout", "10",
+		"--log-level", "ERROR",
+	}
+
+	cmd := exec.CommandContext(ctx, "uvx", args...)
+	cmd.Env = os.Environ()
+
+	stdin, err := cmd.StdinPipe()
+	if err != nil {
+		return nil, fmt.Errorf("stdin pipe: %w", err)
+	}
+	stdout, err := cmd.StdoutPipe()
+	if err != nil {
+		return nil, fmt.Errorf("stdout pipe: %w", err)
+	}
+	cmd.Stderr = os.Stderr
+
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("start: %w", err)
+	}
+
+	c := &linkedInMCPClient{
+		cmd:    cmd,
+		stdin:  bufio.NewWriter(stdin),
+		stdout: bufio.NewScanner(stdout),
+		msgID:  0,
+	}
+	c.stdout.Buffer(make([]byte, 1<<20), 1<<20)
+
+	if err := c.initialize(ctx); err != nil {
+		c.Close()
+		return nil, fmt.Errorf("init: %w", err)
+	}
+	return c, nil
+}
+
+func (c *linkedInMCPClient) nextID() int {
+	c.msgID++
+	return c.msgID
+}
+
+func (c *linkedInMCPClient) initialize(ctx context.Context) error {
+	params := map[string]interface{}{
+		"protocolVersion": "2024-11-05",
+		"capabilities":    map[string]interface{}{},
+		"clientInfo": map[string]string{
+			"name":    "chats-sync",
+			"version": "0.1.0",
+		},
+	}
+	_, err := c.send(ctx, "initialize", params)
+	return err
+}
+
+func (c *linkedInMCPClient) send(ctx context.Context, method string, params interface{}) (json.RawMessage, error) {
+	id := c.nextID()
+	req := map[string]interface{}{
+		"jsonrpc": "2.0",
+		"id":      id,
+		"method":  method,
+	}
+	if params != nil {
+		req["params"] = params
+	}
+	body, err := json.Marshal(req)
+	if err != nil {
+		return nil, fmt.Errorf("marshal: %w", err)
+	}
+
+	if _, err := c.stdin.Write(body); err != nil {
+		return nil, fmt.Errorf("write: %w", err)
+	}
+	if err := c.stdin.WriteByte('\n'); err != nil {
+		return nil, fmt.Errorf("newline: %w", err)
+	}
+	if err := c.stdin.Flush(); err != nil {
+		return nil, fmt.Errorf("flush: %w", err)
+	}
+
+	for c.stdout.Scan() {
+		line := c.stdout.Text()
+		if line == "" {
+			continue
+		}
+		var resp struct {
+			JSONRPC string          `json:"jsonrpc"`
+			ID      int             `json:"id"`
+			Result  json.RawMessage `json:"result,omitempty"`
+			Error   *struct {
+				Code    int    `json:"code"`
+				Message string `json:"message"`
+			} `json:"error,omitempty"`
+		}
+		if err := json.Unmarshal([]byte(line), &resp); err != nil {
+			return nil, fmt.Errorf("unmarshal: %w\nline: %s", err, line[:min(len(line), 500)])
+		}
+		if resp.Error != nil {
+			return nil, fmt.Errorf("rpc error %d: %s", resp.Error.Code, resp.Error.Message)
+		}
+		return resp.Result, nil
+	}
+	return nil, fmt.Errorf("no response: %w", c.stdout.Err())
+}
+
+// msgLimitFor returns the per-conversation message cap for a sync.
+func msgLimitFor(limit int) int {
+	if limit > 0 {
+		return limit
+	}
+	return 100
+}
+
+// wedged reports whether a conversation fetch failure means the MCP server
+// closed its browser and will refuse every later call.
+func wedged(err error) bool {
+	return strings.Contains(err.Error(), "still has a browser open")
+}
+
+// callTool invokes an MCP tool, retrying transient (rate-limit) failures.
+func (c *linkedInMCPClient) callTool(ctx context.Context, name string, params map[string]interface{}) (json.RawMessage, error) {
+	var lastErr error
+	for attempt := 0; attempt < 3; attempt++ {
+		if attempt > 0 {
+			delay := time.Duration(1<<uint(attempt)) * 5 * time.Second
+			select {
+			case <-ctx.Done():
+				return nil, ctx.Err()
+			case <-time.After(delay):
+			}
+		}
+		result, err := c.send(ctx, "tools/call", map[string]interface{}{
+			"name":      name,
+			"arguments": params,
+		})
+		if err == nil {
+			// Tool-level errors surface as a successful RPC with an
+			// isError=true content entry.
+			if hint := toolErrorHint(result); hint != "" {
+				lastErr = fmt.Errorf("%s error: %s", name, hint)
+				if !isTransientLinkedInError(lastErr.Error()) {
+					return nil, lastErr
+				}
+				continue
+			}
+			return result, nil
+		}
+		lastErr = err
+		if !isTransientLinkedInError(err.Error()) {
+			return nil, err
+		}
+	}
+	return nil, fmt.Errorf("%s: %w", name, lastErr)
+}
+
+// toolErrorHint returns the tool's error text when the result has isError set.
+func toolErrorHint(result json.RawMessage) string {
+	var toolRes struct {
+		Content []struct {
+			Type string `json:"type"`
+			Text string `json:"text"`
+		} `json:"content"`
+		IsError bool `json:"isError"`
+	}
+	if err := json.Unmarshal(result, &toolRes); err != nil || !toolRes.IsError {
+		return ""
+	}
+	if len(toolRes.Content) > 0 {
+		return toolRes.Content[0].Text
+	}
+	return "unknown tool error"
+}
+
+// isTransientLinkedInError reports whether a fetch failed due to rate limiting
+// or a transient server error, which may succeed on retry.
+func isTransientLinkedInError(msg string) bool {
+	return strings.Contains(msg, "503") || strings.Contains(msg, "429") ||
+		strings.Contains(msg, "ERR_HTTP_RESPONSE_CODE_FAILURE") ||
+		strings.Contains(msg, "ERR_ABORTED") ||
+		strings.Contains(msg, "Error calling tool") ||
+		strings.Contains(msg, "Unexpected error") ||
+		strings.Contains(msg, "still has a browser open")
+}
+
+func (c *linkedInMCPClient) GetInbox(ctx context.Context, limit int) ([]lnInboxItem, error) {
+	params := map[string]interface{}{
+		"limit": limit,
+	}
+	result, err := c.callTool(ctx, "get_inbox", params)
+	if err != nil {
+		return nil, err
+	}
+
+	var toolRes struct {
+		Content []struct {
+			Type string `json:"type"`
+			Text string `json:"text"`
+		} `json:"content"`
+		IsError bool `json:"isError"`
+	}
+	if err := json.Unmarshal(result, &toolRes); err != nil {
+		return nil, fmt.Errorf("unmarshal tool: %w", err)
+	}
+	if len(toolRes.Content) == 0 {
+		return nil, nil
+	}
+
+	text := toolRes.Content[0].Text
+	return parseLinkedInInbox(text), nil
+}
+
+func (c *linkedInMCPClient) GetConversation(ctx context.Context, username, threadID string, limit int) ([]lnMessage, error) {
+	params := map[string]interface{}{
+		"linkedin_username": username,
+		"thread_id":         threadID,
+		"index":             limit,
+	}
+	result, err := c.callTool(ctx, "get_conversation", params)
+	if err != nil {
+		return nil, err
+	}
+
+	var toolRes struct {
+		Content []struct {
+			Type string `json:"type"`
+			Text string `json:"text"`
+		} `json:"content"`
+		IsError bool `json:"isError"`
+	}
+	if err := json.Unmarshal(result, &toolRes); err != nil {
+		return nil, fmt.Errorf("unmarshal tool: %w", err)
+	}
+	if len(toolRes.Content) == 0 {
+		return nil, nil
+	}
+
+	text := toolRes.Content[0].Text
+	return parseLinkedInConversation(text), nil
+}
+
+func (c *linkedInMCPClient) Close() error {
+	if c.stdin != nil {
+		c.stdin.Flush()
+	}
+	if c.cmd != nil && c.cmd.Process != nil {
+		c.cmd.Process.Kill()
+	}
+	return nil
+}

bin/chats/linkedin_test.go

@@ -0,0 +1,215 @@
+package main
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+func readFixture(t *testing.T, name string) string {
+	t.Helper()
+	data, err := os.ReadFile(filepath.Join("testdata", name))
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(data)
+}
+
+// TestParseLinkedInInbox verifies get_inbox parsing against the v4.22 wire
+// format (testdata/linkedin_inbox.json — synthetic Alice/Bob/Charlie).
+func TestParseLinkedInInbox(t *testing.T) {
+	text := readFixture(t, "linkedin_inbox.json")
+	items := parseLinkedInInbox(text)
+	if len(items) != 2 {
+		t.Fatalf("expected 2 conversations (empty thread url skipped), got %d", len(items))
+	}
+
+	first := items[0]
+	if first.ThreadID == "" {
+		t.Error("expected thread id extracted from reference url")
+	}
+	if first.Participants != "Alice Example" {
+		t.Errorf("participants=%q, want Alice Example", first.Participants)
+	}
+	if !strings.HasPrefix(first.ThreadID, "2-") {
+		t.Errorf("unexpected thread id format %q", first.ThreadID)
+	}
+	if items[1].Participants != "Charlie Example" {
+		t.Errorf("second participant=%q, want Charlie Example", items[1].Participants)
+	}
+}
+
+// TestParseLinkedInInboxBadJSON verifies a non-JSON response yields no items
+// rather than a panic or error.
+func TestParseLinkedInInboxBadJSON(t *testing.T) {
+	if got := parseLinkedInInbox("Session expired"); len(got) != 0 {
+		t.Fatalf("expected no items for non-JSON, got %d", len(got))
+	}
+}
+
+// TestParseLinkedInConversation verifies message extraction from the sections
+// blob (testdata/linkedin_conversation.json — synthetic Alice/Bob).
+func TestParseLinkedInConversation(t *testing.T) {
+	text := readFixture(t, "linkedin_conversation.json")
+	msgs := parseLinkedInConversation(text)
+	if len(msgs) != 2 {
+		t.Fatalf("expected 2 messages, got %d", len(msgs))
+	}
+
+	if msgs[0].From != "Alice Example" {
+		t.Errorf("from=%q, want Alice Example", msgs[0].From)
+	}
+	if !strings.Contains(msgs[0].Text, "Senior Software Engineer") {
+		t.Errorf("alice text missing role, got %q", msgs[0].Text)
+	}
+	if msgs[0].Date == "" {
+		t.Error("expected message date")
+	}
+	if msgs[1].From != "Bob Example" {
+		t.Errorf("from=%q, want Bob Example", msgs[1].From)
+	}
+}
+
+// TestParseLinkedInConversationEmpty verifies empty/non-JSON blobs parse to
+// zero messages.
+func TestParseLinkedInConversationEmpty(t *testing.T) {
+	if got := parseLinkedInConversation("no data here"); len(got) != 0 {
+		t.Fatalf("expected 0 messages, got %d", len(got))
+	}
+}
+
+// TestLinkedInTimestamp verifies weekday+clock resolution to a recent UTC date.
+func TestLinkedInTimestamp(t *testing.T) {
+	// The most recent Wednesday before/equal to "now".
+	ts := linkedInTimestamp("WEDNESDAY", "10:02 AM")
+	parsed, err := time.Parse(time.RFC3339, ts)
+	if err != nil {
+		t.Fatalf("unparseable timestamp %q: %v", ts, err)
+	}
+	if parsed.Weekday() != time.Wednesday {
+		t.Errorf("expected Wednesday, got %s", parsed.Weekday())
+	}
+	if parsed.Hour() != 10 || parsed.Minute() != 2 {
+		t.Errorf("expected 10:02, got %02d:%02d", parsed.Hour(), parsed.Minute())
+	}
+
+	now := time.Now()
+	diff := now.Sub(parsed)
+	if diff < 0 || diff > 7*24*time.Hour {
+		t.Errorf("timestamp %s is not within the last week of %s", parsed, now)
+	}
+
+	if got := linkedInTimestamp("MONDAY", "garbage"); got != "" {
+		t.Errorf("expected empty for bad clock, got %q", got)
+	}
+	if got := linkedInTimestamp("", "1:22 PM"); got != "13:22" {
+		t.Errorf("expected bare 13:22 for missing weekday, got %q", got)
+	}
+
+	// Relative day headers must resolve to full dates, not bare clocks.
+	today := linkedInTimestamp("TODAY", "9:42 AM")
+	yp, err := time.Parse(time.RFC3339, today)
+	if err != nil {
+		t.Fatalf("TODAY unparseable %q: %v", today, err)
+	}
+	if yp.Year() != now.Year() || yp.Month() != now.Month() || yp.Day() != now.Day() {
+		t.Errorf("TODAY expected %v, got %v", now, yp)
+	}
+	yest := linkedInTimestamp("YESTERDAY", "3:00 PM")
+	yp, err = time.Parse(time.RFC3339, yest)
+	if err != nil {
+		t.Fatalf("YESTERDAY unparseable %q: %v", yest, err)
+	}
+	if yp.Day() != now.AddDate(0, 0, -1).Day() {
+		t.Errorf("YESTERDAY expected day %d, got %d", now.AddDate(0, 0, -1).Day(), yp.Day())
+	}
+
+	// MON DD header (e.g. "JUN 25"): must resolve to a full date. The
+	// timestamp should fall within the current year (falling back to the
+	// prior year if the date would be in the future).
+	md := linkedInTimestamp("JUN 25", "10:48 AM")
+	mp, err := time.Parse(time.RFC3339, md)
+	if err != nil {
+		t.Fatalf("MON DD unparseable %q: %v", md, err)
+	}
+	if mp.Year() != now.Year() && mp.Year() != now.Year()-1 {
+		t.Errorf("JUN 25 expected year %d or %d, got %d", now.Year(), now.Year()-1, mp.Year())
+	}
+	if mp.Month() != time.June || mp.Day() != 25 {
+		t.Errorf("JUN 25 expected Jun 25, got %s %d", mp.Month(), mp.Day())
+	}
+	if mp.After(now) {
+		t.Errorf("JUN 25 resolved to the future: %s > %s", mp, now)
+	}
+}
+
+// TestTransientLinkedInError verifies rate-limit errors are retryable but
+// genuine failures are not.
+func TestTransientLinkedInError(t *testing.T) {
+	retryable := []string{
+		"get_conversation error: Error calling tool 'get_conversation'",
+		"get_conversation error: Unexpected error in get_conversation: net::ERR_HTTP_RESPONSE_CODE_FAILURE",
+		"rpc error 503: rate limited",
+		"rpc error 429: too many requests",
+		"get_conversation: get_conversation error: This server still has a browser open on the profile.",
+	}
+	for _, msg := range retryable {
+		if !isTransientLinkedInError(msg) {
+			t.Errorf("expected %q to be transient", msg)
+		}
+	}
+	permanent := []string{
+		"get_inbox error: bad credentials",
+		"rpc error -32602: Invalid request parameters",
+		"unmarshal: unexpected end of JSON input",
+	}
+	for _, msg := range permanent {
+		if isTransientLinkedInError(msg) {
+			t.Errorf("expected %q to be permanent", msg)
+		}
+	}
+}
+
+// TestMsgLimitFor verifies the per-conversation message cap resolution.
+func TestMsgLimitFor(t *testing.T) {
+	if got := msgLimitFor(0); got != 100 {
+		t.Errorf("expected default 100, got %d", got)
+	}
+	if got := msgLimitFor(5); got != 5 {
+		t.Errorf("expected 5, got %d", got)
+	}
+}
+
+// TestWedged verifies the browser-open failure is recognized as a wedge.
+func TestWedged(t *testing.T) {
+	if !wedged(errors.New("get_conversation error: This server still has a browser open on the profile")) {
+		t.Error("expected wedged error to be recognized")
+	}
+	if wedged(errors.New("get_conversation error: bad thing")) {
+		t.Error("unexpected wedge detection")
+	}
+}
+
+// TestThreadIDFromURL verifies thread id extraction.
+func TestThreadIDFromURL(t *testing.T) {
+	cases := []struct {
+		url, want string
+	}{
+		{"/messaging/thread/2-abc123/", "2-abc123"},
+		{"/messaging/thread/2-abc123", "2-abc123"},
+		{"", ""},
+		{"/messaging/thread/", ""},
+	}
+	for _, c := range cases {
+		got := threadIDFromURL(c.url)
+		if c.want == "" && validThreadID(got) {
+			t.Errorf("threadIDFromURL(%q) = %q, want empty", c.url, got)
+		}
+		if c.want != "" && got != c.want {
+			t.Errorf("threadIDFromURL(%q) = %q, want %q", c.url, got, c.want)
+		}
+	}
+}

bin/chats/main.go

@@ -40,8 +40,7 @@ func main() {
 			fmt.Fprintf(os.Stderr, "chats: WhatsApp not implemented yet\n")
 			os.Exit(1)
 		case "linkedin":
-			fmt.Fprintf(os.Stderr, "chats: LinkedIn not implemented yet\n")
+			os.Exit(runSyncLinkedIn(platformArgs))
-			os.Exit(1)
 		default:
 			fmt.Fprintf(os.Stderr, "chats: unknown platform %q\n", platform)
 			os.Exit(2)

bin/chats/refresh-linkedin-session

@@ -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()

bin/chats/sync_linkedin_cmd.go

@@ -0,0 +1,105 @@
+package main
+
+import (
+	"context"
+	"flag"
+	"fmt"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"time"
+)
+
+func checkLinkedInSession(userDataDir string) (bool, error) {
+	// Validate the source-session files without launching a browser. A full
+	// `--status` run spawns Chromium and loads /feed/, doubling the automation
+	// exposed to LinkedIn (429 rate limits) before the sync even starts.
+	root := filepath.Dir(userDataDir)
+	sessionFiles := []string{
+		filepath.Join(root, "source-state.json"),
+		filepath.Join(root, "cookies.json"),
+		filepath.Join(userDataDir, "Default", "Cookies"),
+	}
+	for _, f := range sessionFiles {
+		if _, err := os.Stat(f); err != nil {
+			return true, fmt.Errorf("missing session file %s", f)
+		}
+	}
+	return false, nil
+}
+
+func runSyncLinkedIn(args []string) int {
+	fs := flag.NewFlagSet("chats sync linkedin", flag.ContinueOnError)
+	limit := fs.Int("limit", 0, "max messages per conversation (0 = all)")
+	refresh := fs.Bool("refresh", false, "refresh session from live webtop browser before sync")
+	help := fs.Bool("help", false, "")
+	fs.SetOutput(os.Stderr)
+	if err := fs.Parse(args); err != nil {
+		return 2
+	}
+	if *help {
+		fmt.Fprintln(os.Stderr, "usage: chats sync linkedin [--limit N] [--refresh]")
+		return 0
+	}
+
+	userDataDir := envVar("LINKEDIN_USER_DATA_DIR", "")
+	if userDataDir == "" {
+		home, _ := os.UserHomeDir()
+		userDataDir = home + "/.linkedin-mcp/profile"
+	}
+
+	if *refresh {
+		if code := refreshLinkedInSession(userDataDir); code != 0 {
+			return code
+		}
+	}
+
+	// Check session files first (no browser launch).
+	loginNeeded, err := checkLinkedInSession(userDataDir)
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "chats: linkedin status check: %v\n", err)
+	}
+	if loginNeeded {
+		fmt.Fprintf(os.Stderr, "chats: LinkedIn session missing. Run:\n")
+		fmt.Fprintf(os.Stderr, "  chats sync linkedin --refresh\n")
+		fmt.Fprintf(os.Stderr, "or point LINKEDIN_USER_DATA_DIR at a valid session\n")
+		return 1
+	}
+
+	src := NewLinkedInMCPSource(userDataDir)
+
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Minute)
+	defer cancel()
+
+	start := time.Now()
+	if err := src.Sync(ctx, chatsDir(), *limit); err != nil {
+		fmt.Fprintf(os.Stderr, "chats sync linkedin: %v\n", err)
+		return 1
+	}
+	fmt.Printf("chats sync linkedin: completed in %s\n", time.Since(start).Round(time.Millisecond))
+	return 0
+}
+
+// refreshLinkedInSession re-syncs the LinkedIn source session from the live
+// webtop browser via the vendored refresh-linkedin-session helper.
+func refreshLinkedInSession(userDataDir string) int {
+	exe, err := os.Executable()
+	if err != nil {
+		fmt.Fprintf(os.Stderr, "chats: resolve executable: %v\n", err)
+		return 1
+	}
+	helper := filepath.Join(filepath.Dir(exe), "refresh-linkedin-session")
+	if _, err := os.Stat(helper); err != nil {
+		// Fall back to the source tree helper next to this command file.
+		helper = "bin/chats/refresh-linkedin-session"
+	}
+	root := filepath.Dir(userDataDir)
+	cmd := exec.Command(helper, "--root", root)
+	cmd.Stdout = os.Stderr
+	cmd.Stderr = os.Stderr
+	if err := cmd.Run(); err != nil {
+		fmt.Fprintf(os.Stderr, "chats: linkedin session refresh: %v\n", err)
+		return 1
+	}
+	return 0
+}

bin/chats/testdata/linkedin_conversation.json

@@ -0,0 +1,12 @@
+{
+  "url": "https://www.linkedin.com/messaging/thread/2-YWxpY2UtYm9iLXRocmVhZC0xMjM=/",
+  "sections": {
+    "conversation": "WEDNESDAY\nAlice Example sent the following message at 10:02 AM\nView Alice Example's profile\nAlice Example  (She/Her)  10:02 AM\nHi Bob, we have a Senior Software Engineer role that matches your Go and Python background. Happy to share more if you are open to a chat.\n\nBob Example sent the following messages at 1:22 PM\nView Bob Example's profile\nBob Example  1:22 PM\nHi Alice, thanks for reaching out — yes, I am open to exploring a Senior Software Engineer role. Happy to do a short video call.\n"
+  },
+  "references": {
+    "conversation": [
+      {"kind": "person", "url": "/in/alice-example/", "text": "Alice Example"},
+      {"kind": "person", "url": "/in/bob-example/", "text": "Bob Example"}
+    ]
+  }
+}

bin/chats/testdata/linkedin_inbox.json

@@ -0,0 +1,34 @@
+{
+  "url": "https://www.linkedin.com/messaging/",
+  "sections": {
+    "inbox": "Messaging\nInbox\nConversation List\nAlice Example\nExciting opportunity for a senior software engineer\n"
+  },
+  "references": {
+    "inbox": [
+      {
+        "kind": "conversation",
+        "url": "/messaging/thread/2-YWxpY2UtYm9iLXRocmVhZC0xMjM=/",
+        "context": "inbox",
+        "text": "Alice Example"
+      },
+      {
+        "kind": "person",
+        "url": "/in/alice-example/",
+        "text": "Alice Example",
+        "context": "inbox"
+      },
+      {
+        "kind": "conversation",
+        "url": "/messaging/thread/2-Y2hhcmxpZS1ib2ItdGhyZWFkLTQ1Ng==/",
+        "context": "inbox",
+        "text": "Charlie Example"
+      },
+      {
+        "kind": "conversation",
+        "url": "/messaging/thread/",
+        "context": "inbox",
+        "text": "should-be-skipped"
+      }
+    ]
+  }
+}

bin/doc.go

@@ -0,0 +1,2 @@
+// Deprecated shebang mains at bin root (serve.go is tagged brain_serve).
+package main

bin/kb/search

@@ -1,34 +1,37 @@
 #!/usr/bin/env bash
-# bin/kb/search - Go deduction search over the brain (model served by daemon).
+# bin/kb/search — deprecated wrapper. Use bin/brain/search.go.
-# Builds the kbsearch binary on first run / when source changes, then execs it.
+# Sets CGO for ladybug, builds a binary (embed daemon needs a real executable),
+# then execs it. Prints one deprecation line.
 set -euo pipefail
 
-KB="$(cd "$(dirname "$0")/../.." && pwd)"
+ROOT="$(cd "$(dirname "$0")/../.." && pwd)"
-BIN="$KB/var/bin/kbsearch"
+BIN="$ROOT/var/bin/brain-search"
-SRC="$KB/bin/kbsearch"
+SRC="$ROOT/internal/brain"
+CMD="$ROOT/bin/brain"
 
-mkdir -p "$KB/var/bin"
+mkdir -p "$ROOT/var/bin"
 
-# Rebuild if binary missing or any .go source newer
 need_build=0
 if [ ! -x "$BIN" ]; then
-    need_build=1
+	need_build=1
 else
-    # Check if any .go in kbsearch is newer than binary
+	while IFS= read -r -d '' f; do
-    while IFS= read -r -d '' f; do
+		if [ "$f" -nt "$BIN" ]; then
-        if [ "$f" -nt "$BIN" ]; then
+			need_build=1
-            need_build=1
+			break
-            break
+		fi
-        fi
+	done < <(find "$SRC" "$CMD" -name '*.go' -print0 2>/dev/null)
-    done < <(find "$SRC" -name '*.go' -print0 2>/dev/null)
 fi
 
 if [ "$need_build" -eq 1 ]; then
-    echo "Building kbsearch..." >&2
+	echo "Building brain/search..." >&2
-    (cd "$SRC" && \
+	(
-        CGO_CFLAGS="-I$KB/lib-ladybug" \
+		cd "$ROOT" &&
-        CGO_LDFLAGS="-L$KB/lib-ladybug -Wl,-rpath,$KB/lib-ladybug" \
+			CGO_CFLAGS="-I$ROOT/lib-ladybug" \
-        go build -tags system_ladybug -o "$BIN" .) || exit 1
+			CGO_LDFLAGS="-L$ROOT/lib-ladybug -Wl,-rpath,$ROOT/lib-ladybug" \
+			go build -tags system_ladybug -o "$BIN" ./bin/brain
+	) || exit 1
 fi
 
+echo "bin/kb/search is deprecated; use bin/brain/search.go" >&2
 exec "$BIN" "$@"

bin/kbsearch/go.mod

@@ -1,23 +0,0 @@
-module github.com/eSlider/2dph/bin/kbsearch
-
-go 1.26.0
-
-require (
-	github.com/LadybugDB/go-ladybug v0.17.0
-	github.com/chewxy/math32 v1.11.2
-	github.com/daulet/tokenizers v1.27.0
-)
-
-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
-)

bin/kbsearch/go.sum

@@ -1,44 +0,0 @@
-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/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/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=
-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=

bin/kbsearch/main.go

@@ -1,44 +0,0 @@
-// bin/kbsearch - the Go implementation of bin/kb/search (nested module so the
-// root `go test ./...` and CI never compile it against native ladyships).
-//
-// Usage (built/run by ./bin/kb/search):
-//
-//	kbsearch "query" [--root facts|info] [--repo P] [-n N] [--json]
-//	kbsearch serve [port]           start the embedding daemon
-//	kbsearch --list-model           print the resolved model dir
-//
-// The potion-multilingual model is loaded only in `serve`; a CLI reuses the
-// daemon over localhost HTTP (falling back to in-process embedding).
-package main
-
-import (
-	"fmt"
-	"log"
-	"os"
-	"strconv"
-)
-
-func main() {
-	if len(os.Args) > 1 && os.Args[1] == "serve" {
-		port := 17830
-		if len(os.Args) > 2 {
-			if p, err := strconv.Atoi(os.Args[2]); err == nil {
-				port = p
-			}
-		}
-		if err := serve(port); err != nil {
-			log.Fatalf("kbsearch serve: %v", err)
-		}
-		return
-	}
-	if len(os.Args) > 1 && os.Args[1] == "--list-model" {
-		dir, err := modelDir()
-		if err != nil {
-			fmt.Fprintln(os.Stderr, err)
-			os.Exit(1)
-		}
-		fmt.Println(dir)
-		return
-	}
-	os.Exit(runSearch(os.Args[1:]))
-}

bin/kbsearch/types.go

@@ -1,16 +0,0 @@
-// Common types and helpers for kbsearch.
-package main
-
-import "os"
-
-func eps() string { return os.Getenv("KBTEST_EPS") }
-
-// 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:"-"` // for repo filtering, not in output
-	Score   float64 `json:"score"`
-	Snippet string  `json:"snippet,omitempty"`
-}

bin/mail/sync/cli.go

@@ -32,6 +32,7 @@ func ParseCLI(args []string) (CLIConfig, int, error) {
 		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")
 	)
@@ -60,6 +61,7 @@ func ParseCLI(args []string) (CLIConfig, int, error) {
 		Offset:  *offset,
 		Force:   *force,
 		DryRun:  *dryRun,
+		Query:   *query,
 		Policy:  RetryPolicy{},
 	}
 	cli := CLIConfig{Sync: cfg, Env: *env, Sources: *srcs}
@@ -95,7 +97,7 @@ func Main(args []string) int {
 		return code
 	}
 	if cli.Help {
-		fmt.Fprintln(os.Stderr, "usage: bin/mail/sync.go [--source onlyoffice,gmail] [--limit N] [--offset N] [--workers N] [--force] [--dry-run]")
+		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)

bin/mail/sync/sync.go

@@ -112,6 +112,7 @@ type SyncConfig struct {
 	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
 }
 
@@ -136,9 +137,17 @@ 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   *GmailClient
+	c     gmailAPI
-	cur string
+	cur   string
+	query string
 }
 
 func (s *ooSource) Folder() string      { return "inbox" }
@@ -171,7 +180,11 @@ func (s *ooSource) DownloadAttachment(ctx context.Context, msg *Message, att Att
 }
 
 func (s *gmailSource) ListIDs(ctx context.Context, limit int, cursor string) ([]string, string, error) {
-	ids, next, err := s.c.ListIDs(ctx, "in:inbox", limit, cursor)
+	q := s.query
+	if q == "" {
+		q = "in:inbox"
+	}
+	ids, next, err := s.c.ListIDs(ctx, q, limit, cursor)
 	return ids, next, err
 }
 
@@ -207,7 +220,7 @@ func Run(ctx context.Context, cfg SyncConfig) (*SyncStats, error) {
 		if err != nil {
 			return nil, fmt.Errorf("gmail init: %w", err)
 		}
-		sources = append(sources, &gmailSource{c: gm})
+		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)")

bin/mail/sync/sync_test.go

@@ -221,6 +221,71 @@ func TestCollectParts(t *testing.T) {
 	}
 }
 
+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))
 }

bin/serve.go

@@ -1,27 +1,22 @@
-//usr/bin/env go run "$0" "$@"; exit
+//usr/bin/env go run -tags=brain_serve "$0" "$@"; exit
-// bin/serve.go - async Go HTTP server for the 2dph brain (see bin/server).
+//go:build brain_serve
 //
-//	KB_ROOT=/path/to/2dph ./bin/serve.go        # serve the brain
+// bin/serve.go — deprecated; use bin/brain/serve.go.
-//	KB_SEARCH_CMD=... KB_WORKERS=4 KB_PORT=8630 ./bin/serve.go
-//
-// Shebang trick: the first line is a Go `//` comment; when executed, env runs
-// `go run "$0"` so this file doubles as an executable script. 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 (
+	"fmt"
 	"os"
 
-	"github.com/eSlider/2dph/bin/server"
+	"github.com/eSlider/2dph/internal/httpapi"
 )
 
 func main() {
-	if env := os.Getenv("KB_ROOT"); env == "" {
+	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)
 		}
 	}
-	server.Run()
+	httpapi.Run()
 }

bin/tools/test_bin_layout.py

@@ -0,0 +1,36 @@
+"""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")

bin/tools/test_published_docs.py

@@ -0,0 +1,48 @@
+"""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_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",
+            )

docs/README.md

@@ -6,5 +6,9 @@ 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)
+
+Search: `bin/brain/search.go "query"` (HTTP: `bin/brain/serve.go`). `--hop` is
+not a walk; the flag errors until File/FROM_FILE edges exist.
 
 Published docs live here and mirror the project state.

docs/design.md

@@ -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,
+`--hop` is not implemented yet (needs File/FROM_FILE edges). The flag is an
-`related:` files, vector-neighbour leaves) — the deduction walk.
+error; it is not a graph walk.
 
 ## Who / What / How / Where / When + evidence
 

go.mod

@@ -1,8 +1,25 @@
 module github.com/eSlider/2dph
 
-go 1.25.0
+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
+)

go.sum

@@ -1,14 +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/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
+github.com/chewxy/math32 v1.11.2 h1:IufN08Zwr1NKuWfY+4Tz55BcwKmyKKNdOP7KtumehnM=
-github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
+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/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
+github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
-github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
-github.com/stretchr/testify v1.7.0 h1:nwc3DEeHmmLAfoZucVR881uASk0Mfjw8xYJ99tb5CcY=
+github.com/klauspost/compress v1.18.5 h1:/h1gH5Ce+VWNLSWqPzOVn6XBO+vJbCNGvjoaGBFW2IE=
-github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg=
+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=

go.work

@@ -0,0 +1,3 @@
+go 1.26
+
+use .

internal/brain/db.go

@@ -1,10 +1,12 @@
-// Brain connection management using go-ladybug.
+//go:build cgo && system_ladybug
-package main
+
+package brain
 
 import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"strings"
 
 	lbug "github.com/LadybugDB/go-ladybug"
 )
@@ -44,10 +46,10 @@ func dbPath() string {
 }
 
 func openBrain() error {
-	return openWithOpts(2, eps())
+	return openWithSandbox(eps())
 }
 
-func openWithOpts(allow int, epsv string) error {
+func openWithSandbox(epsv string) error {
 	cfg := lbug.DefaultSystemConfig()
 	cfg.MaxNumThreads = 8
 	cfg.BufferPoolSize = 1 << 30 // 1GB
@@ -57,20 +59,30 @@ func openWithOpts(allow int, epsv string) error {
 	if err != nil {
 		return fmt.Errorf("OpenDatabase: %w", err)
 	}
-	if epsv != "" {
-		if _, err := conn.Query("SET STREAM_SANDBOX = '" + epsv + "'"); err != nil {
-			return 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

internal/brain/doc.go

@@ -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

internal/brain/model.go

@@ -1,9 +1,11 @@
+//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 main
+package brain
 
 import (
 	"encoding/json"

internal/brain/modeldir.go

@@ -1,5 +1,4 @@
-// modelDir returns the resolved potion-multilingual-128m model directory.
+package brain
-package main
 
 import (
 	"fmt"

internal/brain/rank/args.go

@@ -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
+}

internal/brain/rank/query.go

@@ -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"

internal/brain/rank/rank.go

@@ -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
+}

internal/brain/rank/rank_test.go

@@ -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)
+	}
+}

internal/brain/search.go

@@ -1,5 +1,6 @@
-// Hybrid FTS + vector search implementation, plus daemon client/server.
+//go:build cgo && system_ladybug
-package main
+
+package brain
 
 import (
 	"bytes"
@@ -13,12 +14,12 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
-	"sort"
 	"strconv"
-	"strings"
+	"syscall"
 	"time"
 
 	lbug "github.com/LadybugDB/go-ladybug"
+	"github.com/eSlider/2dph/internal/brain/rank"
 )
 
 const defaultPort = 17830
@@ -26,45 +27,15 @@ const daemonPath = "/embed"
 const healthPath = "/health"
 
 func runSearch(args []string) int {
-	// Manual flag parsing to allow flags after query (like Python argparse)
+	opt, err := rank.ParseArgs(args)
-	root := ""
+	if err != nil {
-	repo := ""
+		fmt.Fprintf(os.Stderr, "brain/search: %v\n%s\n", err, rank.Usage)
-	limit := 20
+		return 2
-	jsonOut := false
-	listModel := false
-
-	var queryArgs []string
-	for i := 0; i < len(args); i++ {
-		switch args[i] {
-		case "--root":
-			if i+1 < len(args) {
-				root = args[i+1]
-				i++
-			}
-		case "--repo":
-			if i+1 < len(args) {
-				repo = args[i+1]
-				i++
-			}
-		case "-n":
-			if i+1 < len(args) {
-				if n, err := strconv.Atoi(args[i+1]); err == nil {
-					limit = n
-				}
-				i++
-			}
-		case "--json":
-			jsonOut = true
-		case "--list-model":
-			listModel = true
-		default:
-			if !strings.HasPrefix(args[i], "-") {
-				queryArgs = append(queryArgs, args[i])
-			}
-		}
 	}
+	root, repo, limit, query := opt.Root, opt.Repo, opt.Limit, opt.Query
+	jsonOut := opt.JSONOut
 
-	if listModel {
+	if opt.ListModel {
 		dir, err := modelDir()
 		if err != nil {
 			fmt.Fprintln(os.Stderr, err)
@@ -74,12 +45,6 @@ func runSearch(args []string) int {
 		return 0
 	}
 
-	query := strings.TrimSpace(strings.Join(queryArgs, " "))
-	if query == "" {
-		fmt.Fprintln(os.Stderr, "usage: kbsearch \"query\" [--root facts|info] [--repo REPO] [-n N] [--json]")
-		return 1
-	}
-
 	if err := openBrain(); err != nil {
 		fmt.Fprintf(os.Stderr, "open brain: %v\n", err)
 		return 1
@@ -103,17 +68,7 @@ func runSearch(args []string) int {
 		fmt.Fprintf(os.Stderr, "vec: %v\n", err)
 	}
 
-	results := hybrid(fts, vec, limit)
+	results := rank.RankAndFilter(fts, vec, root, repo, limit)
-
-	if root != "" {
-		results = filterRoot(results, root)
-	}
-	if repo != "" {
-		results = filterRepo(results, repo)
-	}
-	if len(results) > limit {
-		results = results[:limit]
-	}
 
 	for i := range results {
 		if results[i].Text != "" {
@@ -150,10 +105,7 @@ func b2i(err error) int {
 }
 
 func queryFTS(text string, limit int) ([]Hit, error) {
-	stmt, err := conn.Prepare(
+	stmt, err := conn.Prepare(rank.FTSStmt)
-		"CALL QUERY_FTS_INDEX('Leaf', 'id', $q) " +
-			"RETURN node.id, node.text, node.root, node.source, score ORDER BY score LIMIT $n",
-	)
 	if err != nil {
 		return nil, err
 	}
@@ -170,10 +122,7 @@ func queryVector(emb []float64, limit int) ([]Hit, error) {
 	for i, v := range emb {
 		embList[i] = v
 	}
-	stmt, err := conn.Prepare(
+	stmt, err := conn.Prepare(rank.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",
-	)
 	if err != nil {
 		return nil, err
 	}
@@ -215,10 +164,10 @@ func rowsToHits(res *lbug.QueryResult) ([]Hit, error) {
 
 // JSON output types
 type jsonOut struct {
-	Query      string     `json:"query"`
+	Query      string    `json:"query"`
-	RootFilter string     `json:"root_filter"`
+	RootFilter string    `json:"root_filter"`
-	Count      int        `json:"count"`
+	Count      int       `json:"count"`
-	Results    []jsonHit  `json:"results"`
+	Results    []jsonHit `json:"results"`
 }
 
 type jsonHit struct {
@@ -248,70 +197,6 @@ func toJSONOut(hits []Hit, query, rootFilter string) *jsonOut {
 	}
 }
 
-func hybrid(fts, vec []Hit, limit int) []Hit {
-	byID := make(map[string]Hit)
-	rrf := make(map[string]float64)
-
-	for rank, h := range fts {
-		byID[h.ID] = h
-		rrf[h.ID] += 1.0 / (60 + float64(rank+1))
-	}
-	for rank, h := range vec {
-		if _, ok := byID[h.ID]; !ok {
-			byID[h.ID] = h
-		} else {
-			existing := byID[h.ID]
-			if existing.Score == 0 {
-				existing.Score = h.Score
-				byID[h.ID] = existing
-			}
-		}
-		rrf[h.ID] += 1.0 / (60 + float64(rank+1))
-	}
-
-	type scored struct {
-		id  string
-		rrf float64
-	}
-	var scoredList []scored
-	for id, v := range rrf {
-		scoredList = append(scoredList, scored{id, v})
-	}
-	sort.Slice(scoredList, func(i, j int) bool {
-		return scoredList[i].rrf > scoredList[j].rrf
-	})
-
-	var out []Hit
-	for i, s := range scoredList {
-		if i >= limit {
-			break
-		}
-		h := byID[s.id]
-		out = append(out, h)
-	}
-	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
-}
-
 func resultsToDicts(hits []Hit) []any {
 	out := make([]any, len(hits))
 	for i, h := range hits {
@@ -363,7 +248,7 @@ func serve(port int) error {
 	})
 
 	addr := fmt.Sprintf("127.0.0.1:%d", port)
-	log.Printf("kbsearch daemon listening on %s", addr)
+	log.Printf("brain search daemon listening on %s", addr)
 	return http.ListenAndServe(addr, mux)
 }
 
@@ -382,10 +267,16 @@ func embedQuery(text string) ([]float64, error) {
 			port = p
 		}
 	}
-	emb, err := tryDaemon(text, port)
+	if emb, err := tryDaemon(text, port); err == nil {
-	if 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 {
@@ -447,9 +338,13 @@ func ensureDaemon(port int) error {
 	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)
@@ -463,3 +358,21 @@ func ensureDaemon(port int) error {
 	}
 	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)
+}

internal/brain/types.go

@@ -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

internal/brain/yaml.go

@@ -1,5 +1,5 @@
 // YAML emitter ported from bin/kb/yamlout.py — preserves insertion order.
-package main
+package brain
 
 import (
 	"fmt"

internal/httpapi/server.go

@@ -2,15 +2,10 @@
 //
 // Async by design: every request runs on its own goroutine, and CPU-heavy
 // searches are serialized through a bounded worker pool (a counting
-// semaphore) so N requests can't spawn N Python interpreters at once.
+// semaphore) so N requests can't spawn N search processes at once.
 //
-// Used by bin/serve.go which is a self-executing shebang script:
+// Used by bin/brain/serve.go.
-//
+package httpapi
-//	///usr/bin/env go run "$0" "$@"; exit
-//	package main
-//	import "github.com/eSlider/2dph/bin/server"
-//	func main() { server.Run() }
-package server
 
 import (
 	"context"
@@ -100,8 +95,8 @@ func writeRaw(w http.ResponseWriter, code int, body []byte) {
 	w.Write(body)
 }
 
-// brainSearcher shells out to bin/kb/search --json. A single python search
+// brainSearcher shells out to the Go brain-search binary (not Python).
-// is bounded and short-lived; the worker pool keeps at most N live.
+// A single search is bounded and short-lived; the worker pool keeps at most N live.
 type brainSearcher struct {
 	cmdPath string
 	timeout time.Duration
@@ -122,15 +117,18 @@ func (b *brainSearcher) Search(ctx context.Context, query string, limit int) ([]
 	return out, nil
 }
 
-// Run starts the HTTP server. Reads env: KB_SEARCH_CMD (default bin/kb/search,
+func defaultSearchCmd(root string) string {
-// relative to the repo root given by KB_ROOT), KB_WORKERS (default 4), KB_PORT
+	if env := os.Getenv("KB_SEARCH_CMD"); env != "" {
-// (default 8630).
+		return env
+	}
+	return filepath.Join(root, "var", "bin", "brain-search")
+}
+
+// Run starts the HTTP server. Reads env: KB_SEARCH_CMD (default
+// $KB_ROOT/var/bin/brain-search), KB_WORKERS (default 4), KB_PORT (default 8630).
 func Run() {
 	root := os.Getenv("KB_ROOT")
-	searchPath := os.Getenv("KB_SEARCH_CMD")
+	searchPath := defaultSearchCmd(root)
-	if searchPath == "" {
-		searchPath = filepath.Join(root, "bin", "kb", "search")
-	}
 	workers := 4
 	if raw := os.Getenv("KB_WORKERS"); raw != "" {
 		if n, err := strconv.Atoi(raw); err == nil && n > 0 {
@@ -147,7 +145,7 @@ func Run() {
 	searcher := &brainSearcher{cmdPath: searchPath, timeout: 60 * time.Second}
 	handler := NewServer(searcher, workers)
 	addr := "127.0.0.1:" + strconv.Itoa(port)
-	log.Printf("serve: %s (workers=%d)", addr, workers)
+	log.Printf("serve: %s (workers=%d cmd=%s)", addr, workers, searchPath)
 	if err := http.ListenAndServe(addr, handler); err != nil {
 		log.Fatal(err)
 	}

internal/httpapi/server_test.go

@@ -1,10 +1,11 @@
-package server
+package httpapi
 
 import (
 	"context"
 	"encoding/json"
 	"net/http"
 	"net/http/httptest"
+	"strings"
 	"sync"
 	"sync/atomic"
 	"testing"
@@ -141,6 +142,17 @@ func TestSearchRejectsBadLimit(t *testing.T) {
 	}
 }
 
+func TestDefaultSearchCmdIsBrainNotPython(t *testing.T) {
+	t.Setenv("KB_SEARCH_CMD", "")
+	cmd := defaultSearchCmd("/repo")
+	if strings.Contains(strings.ToLower(cmd), "python") {
+		t.Fatalf("search path still python: %s", cmd)
+	}
+	if !strings.Contains(cmd, "brain") {
+		t.Fatalf("search path must be the Go brain binary, got %s", cmd)
+	}
+}
+
 func TestSearchTimeout(t *testing.T) {
 	fs := &fakeSearcher{delay: time.Second}
 	h := NewServer(fs, 1)

skills/diataxis-docs/SKILL.md

@@ -30,12 +30,11 @@ related:
 ---
 ```
 
-`bin/kb/index` reads this. `type` becomes a searchable column and `related`
+`bin/kb/index` reads this. `type` becomes a searchable column. `related:` is
-becomes a graph edge:
+frontmatter for humans; graph hops from it are not implemented yet.
 
 ```bash
-bin/kb/search "deploy" --type howto
+bin/brain/search.go "deploy"
-bin/kb/search "Stecktafel" --hop 1     # follow links and related
 ```
 
 ## Audit checklist
@@ -46,8 +45,7 @@ bin/kb/search "Stecktafel" --hop 1     # follow links and related
    The indexer chunks on H2, so split files also search better.
 3. Is `status: archive` set on anything superseded? Archived files stay indexed
    but stop competing with current ones for a reader's attention.
-4. Does every explanation link the reference it explains, and vice versa? That
+4. Does every explanation link the reference it explains, and vice versa?
-   link is what `--hop 1` walks.
 
 ## Rule
 

skills/kb-search/SKILL.md

@@ -2,9 +2,10 @@
 name: kb-search
 description: >-
   Deduction search over the 2dph brain (Ladybug graph: ops corpus, portfolio,
-  ssh hosts) with bin/kb/search instead of reading files or grepping repos.
+  ssh hosts) with bin/brain/search.go instead of reading files or grepping
-  Use whenever a question starts with "where is", "what runs on", "which file
+  repos. Use whenever a question starts with "where is", "what runs on",
-  describes", "who is", "how is X done", before opening any documentation.
+  "which file describes", "who is", "how is X done", before opening any
+  documentation.
 ---
 
 # kb-search — deduction over facts and info
@@ -22,15 +23,17 @@ second independent source when local roots cannot confirm. An answer is
 `(not confirmed)`.
 
 ```bash
-bin/kb/search "Matrix federation"                # pointers + snippets, YAML
+bin/brain/search.go "Matrix federation"                # pointers + snippets, YAML
-bin/kb/search "what runs on arc-2" --hop 1       # follow graph edges
+bin/brain/search.go "onlyoffice postgres" --root facts # restrict to confirmed
-bin/kb/search "onlyoffice postgres" --root facts # restrict to confirmed
+bin/brain/search.go "where is cs-lexicon" --json | yq '.[].ref'
-bin/kb/search "where is cs-lexicon" --json | yq '.[].ref'
+bin/kb/get <id> --body                                 # full chunk only when needed
-bin/kb/get <id> --body                           # full chunk only when needed
+bin/kb/stats                                           # index health
-bin/kb/stats                                     # index health
+bin/kb/eval                                            # recall@5 >= 0.95 gate
-bin/kb/eval                                      # recall@5 >= 0.95 gate
 ```
 
+`bin/kb/search` is a deprecated wrapper. `--hop` errors (File/FROM_FILE edges
+are not wired yet); do not treat it as a graph walk.
+
 ## Rules
 
 - Search before you read. Never grep a repo for a concept the graph covers.
@@ -38,8 +41,6 @@ bin/kb/eval                                      # recall@5 >= 0.95 gate
   facts first, then info leafs clearly marked `(not confirmed)`.
 - If recall looks wrong, run `bin/kb/eval`; it gates control questions and
   should stay at or above 95% recall@5.
-- `--hop N` follows sibling leaves, owning files, `related:` links and
-  vector-neighbours — that is the deduction walk, not random expansion.
 - Escalate to `web-search` (the `web-search` skill) as the independent second
   source when both local roots cannot confirm; never report an unconfirmed
   single-source local answer as fact.

skills/web-search/SKILL.md

@@ -18,9 +18,10 @@ bin/web/search "postgres partial index" --lang en --fresh year
 
 ## Web or knowledge base
 
-`bin/kb/search` holds our own facts: the ops stack, portfolio, ssh hosts, the
+`bin/brain/search.go` holds our own facts: the ops stack, portfolio, ssh hosts,
-lexicon. Go there first. Reach for `bin/web/search` when the answer is outside
+the lexicon. Go there first. Reach for `bin/web/search` when the answer is
-our repos: upstream library behaviour, vendor documentation, public standards.
+outside our repos: upstream library behaviour, vendor documentation, public
+standards.
 
 Keep the two apart. A finding is stronger when the reader can see that one
 source was ours and one was not.