docs: update for v0.57.0 (artifact-aware file search) and v0.58.0 (semantic session search)

- AGENTS.md: bump version to v0.58.8, add v0.57.0/v0.58.0 feature sections
- SESSIONS.md: document session_search tool, two-tier pipeline, vector index maintenance
- TELEGRAM.md: /now archives sessions instead of deleting, add session recall docs
- DEVELOPMENT.md: add generate-changelog.sh usage instructions
- CLI.md/WEBUI.md/SANDBOXING.md: KODE_* -> ODEK_*, kode.json -> odek.json fixes
- CHEATSHEET.md: add session_search to tool table

Author: molty3000

AGENTS.md

@@ -12,7 +12,7 @@ It provides context about the project's architecture, conventions, and how to up
 - **Binary:** `odek` — single static binary, ~12 MB, instant startup.
 - **Config:** Five-layer priority: `~/.odek/secrets.env` → `~/.odek/config.json` → `./odek.json` → `ODEK_*` env vars → CLI flags.
 - **Benchmark:** AIEB v2.0 — 80.3% (highest published agent score on the Autonomous Intelligence Engineering Benchmark).
-- **Version:** v0.56.2 — see latest tag at https://github.com/BackendStack21/odek/releases
+- **Version:** v0.58.8 — see latest tag at https://github.com/BackendStack21/odek/releases
 
 ## Source Layout
 
@@ -71,6 +71,8 @@ ReAct cycle: observe → think → act → repeat.
 - **Interaction modes** — engaging (narrated), enhance (persistent), verbose (raw), off.
 - Max 300 iterations by default.
 - **Post-response async processing** (v0.56.0) — skill learning and episode extraction run in background goroutines, eliminating the 2-5 second hang after every `odek run`.
+- **Artifact-aware file search** (v0.57.0) — `search_files` and `multi_grep` skip build/artifact directories (`node_modules`, `vendor`, `.git`, `__pycache__`, `.venv`, etc.) automatically, reducing noise and speeding scans.
+- **Semantic session search** (v0.58.0) — the `session_search` tool uses go-vector RandomProjections + k-NN for semantic similarity search through session content. Finds relevant past conversations even when keywords don't match. Features a two-tier pipeline: vector index (fast, ~1ms) → deepSearch fallback (exhaustive, slower).
 
 ### Tools
 All built-in tools with zero subprocess forks: batch_read, batch_patch, parallel_shell, http_batch, math_eval, diff, count_lines, multi_grep, json_query, tree, checksum, sort, head_tail, base64, tr, word_count, transcribe, browser, read_file, write_file, search_files, patch, shell, delegate_tasks, session_search.

docs/CHEATSHEET.md

@@ -277,6 +277,7 @@ odek mcp                                    # stdio transport
 |------|-------------|
 | `delegate_tasks` | Spawn sub-agent OS processes |
 | `memory` | Persistent fact CRUD with cosine-merge |
+| `session_search` | Browse, search, and recall past sessions (semantic vector search) |
 | `clarify` | Ask the user for clarification |
 | `send_message` | Send text/photo/document to Telegram |
 | `skill_load/list/save/patch/delete` | Skill CRUD |

docs/CLI.md

@@ -342,7 +342,7 @@ Config sources from lowest to highest priority:
 ```
 1.  ~/.odek/config.json    ← Global defaults
 2.  ./odek.json           ← Project overrides
-3.  KODE_* env vars       ← Runtime overrides
+3.  ODEK_* env vars       ← Runtime overrides
 4.  CLI flags             ← Explicit invocation (highest)
 ```
 

docs/DEVELOPMENT.md

@@ -1,5 +1,34 @@
 # Development
 
+## Generate Changelog
+
+CHANGELOG.md is auto-generated from conventional git commits. **Do not edit manually** — edits will be overwritten.
+
+```bash
+# Generate changelog for the next patch release (stdout)
+./generate-changelog.sh --bump patch
+
+# Prepend to CHANGELOG.md
+./generate-changelog.sh --bump patch --prepend
+
+# Release notes for gh release create
+./generate-changelog.sh --bump patch --notes > /tmp/notes.md
+gh release create v0.59.0 --notes-file /tmp/notes.md
+
+# Explicit range
+./generate-changelog.sh --from v0.57.0 --to v0.58.0
+```
+
+Commit prefixes and their section mapping:
+- `feat|feature:` → ### Features
+- `fix|bugfix:` → ### Bug Fixes
+- `perf:` → ### Performance
+- `refactor:` → ### Refactoring
+- `docs:` → ### Documentation
+- `test|tests:` → ### Testing
+- `chore|build|ci:` → ### Infrastructure
+- (unmatched) → ### Other Changes
+
 ## Prerequisites
 
 - Go 1.24+
@@ -94,7 +123,7 @@ cmd/odek/
   wsapprover.go               WSApprover — WebSocket-based approval for serve mode
   subagent_test.go            Tests (flag parsing, JSON stdout, exit codes, tool schema)
   subagent_contract_test.go   Contract tests (flag parsing, stdout protocol, exit codes)
-  subagent_e2e_test.go        E2E tests (16 — KODE_E2E=true, real subprocess spawning)
+  subagent_e2e_test.go        E2E tests (16 — ODEK_E2E=true, real subprocess spawning)
   ui/
     index.html                Single-page web UI (~770 LOC, vanilla JS + CSS)
 docs/                         Documentation

docs/SANDBOXING.md

@@ -23,7 +23,7 @@ odek run --sandbox "protoc --version"
 
 ## Config reference
 
-All sandbox settings are available in `~/.odek/config.json`, `./kode.json`, `ODEK_*` env vars, and CLI flags, following the same [priority chain](CONFIG.md).
+All sandbox settings are available in `~/.odek/config.json`, `./odek.json`, `ODEK_*` env vars, and CLI flags, following the same [priority chain](CONFIG.md).
 
 ### Config file fields
 

docs/SESSIONS.md

@@ -77,6 +77,28 @@ odek session cleanup 0
 # → Cleaned up 12 session(s) older than 0 days.
 ```
 
+**Vector index maintenance:** `odek session cleanup` also removes stale entries from the vector index — orphaned vectors no longer accumulate in `vectors.gob`. This was a bug fix in v0.58.2: the primary cleanup path previously bypassed `Vec.Remove()`, leaving orphaned vectors that silently accumulated.
+
+## Session Search Tool
+
+The `session_search` tool (available inside the agent loop, not as a CLI command) lets the agent browse, search, and recall past sessions by semantic content. This is the primary mechanism for revisiting historical conversations.
+
+| Action | Description |
+|--------|-------------|
+| `list` | Recent sessions (metadata only: ID, task, turns, timestamps, model) |
+| `search` | Semantic keyword search through full message content using vector similarity |
+| `get` | Full session by ID — returns ALL messages including tool calls and results |
+| `find` | Find sessions by task/title |
+
+**How search works — two-tier pipeline:**
+
+1. **Vector index** (fast path) — go-vector RandomProjections embed session summaries into 64-dimensional vectors. Cosine similarity >0.7 returns results in ~1ms. Zero LLM calls.
+2. **DeepSearch** (fallback) — when vector results are insufficient (<2 distinct token matches), falls back to exhaustive text search through session files on disk. Requires 2+ distinct token matches to qualify.
+
+**IMPORTANT:** After `search` returns matching session IDs, use `get` (not `search`) to read the actual conversation content. `get` returns the full `session_messages` array with every user and assistant message.
+
+From inside the Telegram bot, session recall is seamless: the current user message is persisted to the session store *before* the agent loop runs, so `session_search` can find the current conversation's data during the same turn.
+
 ## Programmatic API
 
 ```go

docs/TELEGRAM.md

@@ -197,7 +197,7 @@ The handler uses `sync.Map` for `TelegramApprover` instances, keyed by `chatID`.
 |---|---|
 | `/start` | Welcome message and bot introduction |
 | `/help` | Show all available commands with descriptions |
-| `/new` | Reset the current conversation (clear context) |
+| `/new` | Archive the current session and start a fresh conversation. Archived sessions are timestamped (`tg-<chatID>-<YYYYMMDD>-<HHMMSS>`) and remain visible via `odek session list` |
 | `/stats` | Show session statistics (turn count, model used, etc.) |
 | `/stop` | Cancel a running agent task |
 | `/mode` | Show current agent modes (interaction_mode, sandbox, skills) |
@@ -229,7 +229,9 @@ The `SessionManager` manages per-chat Telegram agent conversations, backed by th
 2. Sessions are persisted to `~/.odek/sessions/tg-<chatID>.json`
 3. An in-memory cache (`map[int64]*ChatSession`) avoids redundant disk reads
 4. Session TTL (default 24h) controls inactivity timeout
-5. Active sessions survive bot restarts — on reconnect, the session is loaded from disk
+5. **Session recall** — the user message is saved to the session store *before* the agent loop runs, enabling `session_search` to find the current conversation's data during the same turn
+6. Active sessions survive bot restarts — on reconnect, the session is loaded from disk
+7. **/now archives** — using `/new` archives the current session with a timestamped ID (`tg-<chatID>-<YYYYMMDD>-<HHMMSS>`) before starting fresh. Archived sessions remain on disk and are visible via `odek session list`.
 
 ### Key Methods
 

docs/WEBUI.md

@@ -171,5 +171,5 @@ Example event sequence:
 ## Tips
 
 - **Security sandbox**: `odek serve --addr localhost:8080` restricts to localhost. Use a reverse proxy (Caddy, nginx) for remote access.
-- **Config inheritance**: `odek serve` reads the same config chain (`~/.odek/config.json` → `./kode.json` → env vars) as `odek run`. Set your model, API key, and sandbox settings there.
+- **Config inheritance**: `odek serve` reads the same config chain (`~/.odek/config.json` → `./odek.json` → env vars) as `odek run`. Set your model, API key, and sandbox settings there.
 - **Session discovery**: reference any saved session via `@sess:ID` in your prompt to give the agent full context from previous conversations.