@pcircle/memesh 4.4.0 → 4.5.1

package/README.ko.md

@@ -117,7 +117,7 @@ Claude Code를 사용한다면 CLI 안에서 MeMesh를 플러그인으로 설치
 /plugin install memesh@pcircle-memesh
 ```
 
-Claude Code가 훅, 스킬, MCP 서버를 자동으로 연결합니다. 세션 내 자동 캡처, 능동적 회상, Claude Code 대화 내 `/memesh` 스킬(remember / recall / learn / forget), 그리고 에이전트가 사용할 수 있는 MCP 도구 `remember` / `recall` / `forget` / `learn`을 모두 얻을 수 있습니다. CLI와 로컬 대시보드도 추가 글로벌 설치 없이 완전히 접근 가능합니다 — `npx @pcircle/memesh <command>`로 모든 CLI 명령을 실행하고, `npx @pcircle/memesh`로 `localhost:3737`의 대시보드를 시작합니다. MCP 서버는 플러그인에 번들된 컴파일 결과물에서 직접 실행됩니다 — `npx` 조회, `npm install -g`, 빌드 단계가 모두 필요 없습니다. 첫 시작 시 `better-sqlite3` 네이티브 바인딩이 없는 경우(예: Node 메이저 버전 업그레이드 이후), 런처가 프로세스 내에서 자동으로 재빌드한 뒤 실행을 이어갑니다.
+Claude Code가 훅, 스킬, MCP 서버를 자동으로 연결합니다. 세션 내 자동 캡처, 능동적 회상, Claude Code 대화 내 `/memesh` 스킬(remember / recall / learn / forget), 그리고 에이전트가 사용할 수 있는 MCP 도구 `remember` / `recall` / `forget` / `learn`을 모두 얻을 수 있습니다. CLI와 로컬 대시보드도 추가 글로벌 설치 없이 완전히 접근 가능합니다 — `npx @pcircle/memesh <command>`로 모든 CLI 명령을 실행하고, `npx @pcircle/memesh`로 `localhost:3737`의 대시보드를 시작합니다. MCP 서버는 플러그인에 번들된 컴파일 결과물에서 직접 실행됩니다 — `npx` 조회, `npm install -g`, 빌드 단계가 모두 필요 없습니다. memesh는 Node에 내장된 `node:sqlite`(22.13+)에 데이터를 저장하므로, Node를 업그레이드해도 잘못된 런타임용으로 빌드된 바이너리가 남지 않습니다.
 
 ### 옵션 B — npm 글로벌 (선택적 최적화)
 
@@ -128,8 +128,8 @@ npm install -g @pcircle/memesh
 ```
 
 > **첫 설치 안내(일회성):**
-> - **네이티브 모듈** — `better-sqlite3`와 `sqlite-vec`는 macOS(arm64/x64), Linux(x64/arm64), Windows x64에서 사전 빌드 바이너리로 설치됩니다. 흔치 않은 플랫폼이거나 사전 빌드가 실패하는 경우 작동하는 C/C++ 툴체인이 필요합니다.
-> - **임베딩 모델** — 로컬 임베딩을 트리거하는 첫 호출(예: 시맨틱 모드의 `recall`)이 `Xenova/all-MiniLM-L6-v2`(~80 MB)를 `~/.memesh/models/`에 다운로드합니다. 이후 호출은 즉시 실행됩니다. 기본 검색 경로(FTS5)는 이 다운로드가 필요하지 않습니다.
+> - **컴파일러가 필요 없습니다** — 데이터베이스 엔진은 Node 자체의 `node:sqlite`입니다. 의미 기반 검색을 담당하는 `sqlite-vec`는 macOS(arm64/x64), Linux(x64/arm64), Windows x64용 사전 빌드 파일로 제공됩니다. 그 외 플랫폼에서는 그냥 없으며, 회상은 키워드 검색으로 유지됩니다. 여기에는 설치 스크립트를 실행하는 것이 전혀 없으므로 `npm install --ignore-scripts`로도 완전히 동작하는 memesh가 설치됩니다.
+> - **시맨틱 검색은 선택 사항** — 기본 검색 경로는 키워드 검색(FTS5)으로, 모델도 다운로드도 필요 없습니다. 의미 기반 검색에는 임베더가 필요합니다: 로컬에서 [Ollama](https://ollama.com)를 실행하거나 클라우드 임베더를 구성하세요(아래 "임베딩" 참조). 없으면 memesh는 키워드 검색만 사용합니다.
 
 ### 1.5단계: MeMesh를 Claude Code에 연결 (npm 경로만)
 
@@ -261,13 +261,12 @@ memesh export-schema \
 
 ## Claude Code에서 자동으로 일어나는 일
 
-모든 것을 수동으로 기억할 필요는 없습니다. MeMesh는 작업 중에 지식을 캡처하고 주입하는 **7가지 훅**이 있습니다:
+모든 것을 수동으로 기억할 필요는 없습니다. MeMesh는 작업 중에 지식을 캡처하고 주입하는 **6가지 훅**이 있습니다:
 
 | 시점 | MeMesh가 수행하는 작업 |
 |------|---|
 | **매 세션 시작** | 가장 관련 있는 메모리 + 과거 교훈의 사전 경고 로드 |
 | **파일 편집 전** | Claude가 코드를 작성하기 전에 파일 또는 프로젝트와 연결된 메모리 회상 |
-| **Bash 명령 전** | (옵트인) Claude가 높은 검증 가능성의 명령어(테스트, 빌드, 린트, 마이그레이션, 배포, 벤치마크)를 백그라운드 에이전트로 실행하도록 유도 |
 | **기억 요청 시** | "remember this" / "guardar en memesh" / "sauvegarder dans memesh" / "記下來" 의도(5개 언어)를 감지하고 Claude가 memesh를 사용하도록 알림 |
 | **모든 `git commit` 후** | 변경 사항을 diff 통계와 함께 기록 |
 | **Claude가 멈출 때** | 편집된 파일, 수정된 에러, 실패로부터 자동 생성된 구조화된 교훈 캡처 |
@@ -285,8 +284,7 @@ memesh export-schema \
 |---|---|---|
 | `MEMESH_DB_PATH` | `~/.memesh/knowledge-graph.db` | SQLite 데이터베이스 위치를 재정의합니다. |
 | `MEMESH_AUTO_CAPTURE` | `true` | 자동 캡처 훅(`Stop`, `PreCompact`)을 완전히 비활성화합니다. |
-| `MEMESH_AUTO_DETECT_LLM` | 미설정(자동 감지 **켜짐**) | `0`으로 설정하면 memesh가 셸 환경에서 발견한 API 키를 사용하지 않습니다. 기본적으로 `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST`가 설정되어 있고 `~/.memesh/config.json`에 프로바이더를 구성하지 않았다면, memesh는 쓰기 측 LLM 기능(통합, 교훈 추출, 자동 태깅, dream)에 이를 사용합니다. 임베딩은 영향을 받지 않습니다 — `embedder.provider`를 명시적으로 설정하지 않는 한 로컬 ONNX(384차원)로 유지됩니다. |
-| `MEMESH_ENABLE_AGENTIC_ORCHESTRATION` | unset | `1`로 설정하면 실험적 작업 모델 프로토콜(CTO / Orchestrator / Agents 프레이밍)을 활성화합니다. 세션 시작 배너, Bash 명령 nudge, `verify_agent_work` 텔레메트리를 추가합니다. 이 프로토콜의 효과는 측정 중이며 아직 입증되지 않았습니다 — 참여하려면 옵트인하세요. **기본값은 OFF**: 코어 메모리 기능은 이 플래그 없이도 작동합니다. |
+| `MEMESH_AUTO_DETECT_LLM` | 미설정(자동 감지 **켜짐**) | `0`으로 설정하면 memesh가 셸 환경에서 발견한 API 키를 사용하지 않습니다. 기본적으로 `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST`가 설정되어 있고 `~/.memesh/config.json`에 프로바이더를 구성하지 않았다면, memesh는 쓰기 측 LLM 기능(통합, 교훈 추출, 자동 태깅, dream)에 이를 사용합니다. 임베딩은 영향을 받지 않습니다 — `embedder.provider`를 `ollama` 또는 `openai`로 명시하지 않는 한 키워드 전용(FTS5)으로 유지됩니다. |
 | `MEMESH_AUTO_UPDATE` | `off` | 자동 업데이트 정책. `off`(기본값)는 자동 업데이트하지 않습니다; `patch`는 `X.Y.Z → X.Y.Z+N`을 허용합니다; `minor`는 `X.Y.Z → X.Y+1.0`을 추가합니다; `major`는 모든 bump를 허용합니다. 허용된 경우, 분리된 `npm install -g`가 세션 종료 시(Stop 훅) 실행되어 작업을 차단하지 않습니다 — 결과는 `~/.memesh/auto-update.log`에 기록됩니다. `~/.memesh/config.json`에서도 `autoUpdate`로 설정 가능합니다(env가 우선). 설치된 버전이 메인테이너에 의해 deprecated된 경우(보안 권고), `off`에서도 `patch`가 강제 허용됩니다 — minor / major bump는 조용한 동작 변화를 피하기 위해 수동으로 유지됩니다. |
 | `OPENAI_API_KEY` | 미설정 | OpenAI 키. `MEMESH_AUTO_DETECT_LLM=0`을 설정하거나 프로바이더를 명시적으로 구성하지 않는 한 LLM 기능에 자동으로 사용됩니다. |
 | `OLLAMA_HOST` | `http://localhost:11434` | 로컬 Ollama 프로바이더를 사용할 때 Ollama 엔드포인트를 재정의합니다. |
@@ -295,6 +293,8 @@ memesh export-schema \
 
 npm이 설치된 버전을 deprecated로 플래그하면(일반적으로 보안 권고), 다음 세션 시작 시 강력한 `⚠️ MeMesh <ver> is DEPRECATED` 배너가 앞에 추가되고, 업그레이드할 때까지 `memesh update-status`가 동일한 라인을 표시합니다. 일시적인 네트워크 실패가 경고를 흐리지 않도록 검사가 `~/.memesh/update-check.<version>.json`에 캐시됩니다.
 
+**폴백 LLM 제공자(Smart Mode).** dashboard의 **Settings → “Fallback providers”**에서 순서가 있는 페일오버 체인을 설정할 수 있습니다 — 기본 제공자가 다운되면 memesh가 목록의 다음 것을 차례로 시도합니다. 로컬 [Ollama](https://ollama.com) 폴백이나 클라우드(OpenAI / Anthropic, API 키 필요)를 추가하세요. 프라이버시 트레이드오프: 클라우드 폴백이 사용되면 메모리 텍스트(비공개일 수 있음)가 해당 제공자로 전송되므로, 프라이버시를 위해 로컬 전용으로 운영한다면 유의하세요.
+
 ---
 
 ## 대시보드
@@ -316,7 +316,7 @@ npm이 설치된 버전을 deprecated로 플래그하면(일반적으로 보안
 
 ## 스마트 기능
 
-**🧠 스마트 검색** — FTS5 + sqlite-vec를 사용해 모든 메모리에서 즉시 검색. 핫 패스에 LLM이 없어 LongMemEval-S에서 R@5 95.60% 달성.
+**🧠 스마트 검색** — "login security"를 검색하면 "OAuth PKCE"에 대한 메모리를 찾습니다. MeMesh는 핫 패스에서 FTS5 + sqlite-vec를 사용하며(LLM-free), 벡터 보완이 관련된 표현까지 도달합니다.
 
 **🌏 띄어쓰기를 하지 않는 문자 검색** — 중국어, 일본어, 한국어, 태국어, 라오어, 크메르어, 반각 가타카나는 인접한 두 글자 묶음으로 색인됩니다. 따라서 「資料庫遷移前一定要先備份」으로 저장한 기억은 전체 문장을 그대로 입력하지 않아도 「備份」으로 찾을 수 있습니다. 저장할 때와 검색할 때 모두 NFC 정규화를 거치므로, macOS나 한국어·베트남어 IME로 입력한 기억도 어느 쪽 표기로든 찾을 수 있습니다.
 
@@ -361,16 +361,18 @@ memesh config set llm.api-key sk-ant-...
 memesh serve  # 대시보드 열기 → Settings 탭
 ```
 
+**과거 세션을 메모리로 캐내기.** `memesh dream run --from-transcripts`는 이 프로젝트의 Claude Code 세션 기록을 읽고, 대화에 묻힌 결정과 교훈을 LLM에게 물어 제안으로 스테이징합니다 — 지식 그래프에는 자동으로 들어가지 않습니다. `memesh dream show <id>`로 하나씩 검토하고 남길 가치가 있는 것을 accept하세요.
+
 ### 자체 임베딩 사용 (선택)
 
-임베딩은 기본적으로 로컬 ONNX 모델(`Xenova/all-MiniLM-L6-v2`, 384차원)을 사용합니다 — API 키 불필요, 데이터가 기기를 벗어나지 않으며, 기본 FTS5 리콜은 아예 필요하지 않습니다. 호스팅형 또는 로컬 서버 임베더를 쓰려면:
+기본적으로 MeMesh는 **키워드 전용** 리콜(FTS5)을 수행합니다 — API 키 불필요, 모델 다운로드 불필요, 데이터가 기기를 벗어나지 않습니다. 시맨틱(의미 기반) 검색은 선택 사항이며 임베더가 필요합니다. 하나를 구성하세요:
 
 ```bash
 memesh config set embedder.provider openai          # or: ollama
 memesh config set embedder.model text-embedding-3-small
 ```
 
-임베더는 **채팅 LLM과 독립적으로** 구성됩니다 — `llm.provider`를 바꿔도 임베딩이 조용히 바뀌지 않습니다. 다른 차원(예: 384 → 1536)으로 전환하면 MeMesh가 다음 쓰기 시 벡터 인덱스를 자동으로 재구축합니다. 지원되는 `embedder.provider`: `onnx`(기본, 로컬), `openai`, `ollama`.
+임베더는 **채팅 LLM과 독립적으로** 구성됩니다 — `llm.provider`를 바꿔도 임베딩이 조용히 바뀌지 않습니다. 다른 차원(예: 768 → 1536)으로 전환하면 MeMesh가 다음 쓰기 시 벡터 인덱스를 자동으로 재구축합니다. 지원되는 `embedder.provider`: `ollama`(로컬), `openai`(호스팅형). 둘 다 없으면 리콜은 키워드 검색으로 유지됩니다.
 
 | | Level 0 (기본) | Level 1 (스마트 모드) |
 |---|---|---|
@@ -383,7 +385,7 @@ memesh config set embedder.model text-embedding-3-small
 
 ---
 
-## 9가지 메모리 도구 전체
+## 7가지 메모리 도구 전체
 
 | 도구 | 역할 |
 |---|---|
@@ -394,7 +396,6 @@ memesh config set embedder.model text-embedding-3-small
 | `import` | 병합 전략(스킵/덮어쓰기/추가)이 있는 메모리 임포트 |
 | `learn` | 실수로부터 구조화된 교훈 기록(에러, 근본 원인, 수정, 예방) |
 | `user_patterns` | 작업 패턴 분석 — 일정, 도구, 강점, 학습 영역 |
-| `verify_agent_work` | 백그라운드 에이전트 작업에 대한 검증 보고서 유지; 청구된 파일 변경사항을 `git diff`와 현실 확인 |
 
 ---
 
@@ -403,7 +404,7 @@ memesh config set embedder.model text-embedding-3-small
 ```
                     ┌─────────────────┐
                     │   Core Engine   │
-                    │  (8 operations) │
+                    │  (7 operations) │
                     └────────┬────────┘
            ┌─────────────────┼─────────────────┐
            │                 │                 │

package/README.md

@@ -84,7 +84,7 @@ flowchart TB
 | Get auto-capture (sessions → lessons → recall) in Claude Code | Path A (plugin) |
 | Run `memesh remember` / `memesh recall` / `memesh doctor` in any terminal | Path B (npm-global) |
 | Open the local dashboard via `memesh serve` (no `npx` lookup delay) | Path B (npm-global) |
-| Plug `memesh-mcp` into Cursor, Cline, or another MCP client | Path B (npm-global) |
+| Plug `memesh-mcp` into Codex CLI, Gemini CLI, Cursor, or another MCP client | Path B (npm-global) |
 | All of the above | **Install both** — they don't conflict |
 
 ### ⚠️ Installing the plugin does NOT install the CLI
@@ -117,21 +117,21 @@ If you use Claude Code, install MeMesh as a plugin from inside the CLI:
 
 Claude Code wires hooks, skills, and the MCP server automatically. You get in-session auto-capture, proactive recall, the `/memesh` skill (remember / recall / learn / forget) inside the Claude Code conversation, and `remember` / `recall` / `forget` / `learn` available as MCP tools to the agent.
 
-The MCP server runs directly from the plugin's bundled compiled output — no `npx` lookup, no build step needed. If the native `better-sqlite3` binding is missing on first start (e.g. after a Node major upgrade), the launcher self-heals by rebuilding it in-process.
+The MCP server runs directly from the plugin's bundled compiled output — no `npx` lookup, no build step, and nothing to compile. memesh stores its data through `node:sqlite`, which is part of Node itself (22.13+), so a Node upgrade cannot leave it with a binary built for the wrong runtime.
 
 > **This installs the plugin only.** You can run CLI commands via `npx @pcircle/memesh <command>` if you absolutely don't want a global install, but typing plain `memesh` in a terminal will report `command not found`. To get a real shell `memesh` command, also run **Option B** below — both paths coexist and share the same memory database. The "Install paths at a glance" diagram above covers this.
 
 ### Option B — npm global (optional optimisation)
 
-If you want the binary directly on your shell `PATH` (so plain `memesh`, `memesh-mcp`, etc. work in any terminal without the per-call `npx` lookup), or you want to expose `memesh-mcp` as a fixed-path stdio command to **non-Claude-Code MCP clients** (Cursor, Cline, terminal-only flows):
+If you want the binary directly on your shell `PATH` (so plain `memesh`, `memesh-mcp`, etc. work in any terminal without the per-call `npx` lookup), or you want to expose `memesh-mcp` as a fixed-path stdio command to **non-Claude-Code MCP clients** (Codex CLI, Gemini CLI, Cursor, Cline, terminal-only flows):
 
 ```bash
 npm install -g @pcircle/memesh
 ```
 
 > **First-install notes (one-time):**
-> - **Native modules** — `better-sqlite3` and `sqlite-vec` install via prebuilt binaries on macOS (arm64/x64), Linux (x64/arm64), and Windows x64. On uncommon platforms or when prebuilds fail, you'll need a working C/C++ toolchain.
-> - **Embedding model** — the first call that triggers a local embedding (e.g. `recall` with semantic mode) downloads `Xenova/all-MiniLM-L6-v2` (~80 MB) into `~/.memesh/models/`. Subsequent calls are instant. The default retrieval path (FTS5) does not require this download.
+> - **No compiler needed** — the database engine is Node's own `node:sqlite`. `sqlite-vec`, which adds meaning-based search, ships as a prebuilt file for macOS (arm64/x64), Linux (x64/arm64) and Windows x64; on any other platform it is simply absent and recall stays on keyword search. Nothing here runs an install script, so `npm install --ignore-scripts` installs a fully working memesh.
+> - **Semantic (meaning-based) search is optional** — the default recall path is FTS5 keyword search, which needs no model and no download. Meaning-based search needs an embedder: run [Ollama](https://ollama.com) locally, or configure a cloud embedder (see "Bring-your-own embeddings" below). Without one, memesh uses keyword search only.
 
 ### Step 1.5: Wire MeMesh into Claude Code (npm path only)
 
@@ -146,6 +146,27 @@ memesh doctor                # verifies "Hooks wired into Claude Code" passes
 
 The hooks coexist with any custom hooks you already have under `~/.claude/hooks/` — `install-hooks` writes additive entries and never overwrites yours. To remove later: `memesh uninstall-hooks`.
 
+### Same memory from Codex CLI and Gemini CLI
+
+`memesh-mcp` is a plain stdio MCP server, so any MCP-capable host can talk to it — not just Claude Code. With Option B installed (`memesh-mcp` on your `PATH`), register it once per host:
+
+```bash
+# OpenAI Codex CLI — writes [mcp_servers.memesh] into ~/.codex/config.toml
+codex mcp add memesh -- memesh-mcp
+
+# Google Gemini CLI — user scope, so it works in every folder
+gemini mcp add -s user memesh memesh-mcp
+```
+
+Every host reads and writes the same `~/.memesh/knowledge-graph.db`, so a memory stored from a Claude Code session is recallable from Codex or Gemini, and the other way around. Verify from either host by asking it to call the `recall` tool, or from a terminal:
+
+```bash
+codex mcp list       # memesh should be listed as enabled
+gemini mcp list      # memesh should show "Connected"
+```
+
+> **Use `memesh-mcp`, not `npx -p @pcircle/memesh`, as the configured command.** `npx -p` resolves to the *local* package whenever the host's working directory is inside a checkout of this repository, silently running whatever state that working tree is in instead of the installed release.
+
 ### Step 2: Store a decision
 
 > The bash examples below assume `memesh` is on your `PATH` (Option B). Option A (plugin-only) users have two equivalent paths: ask in the Claude Code conversation (the `/memesh` skill + MCP tools cover the same flows), or replace `memesh` with `npx @pcircle/memesh` in any shell — same flags, no global install needed.
@@ -263,13 +284,12 @@ Paste tools into any API call
 
 ## What Happens Automatically In Claude Code
 
-You don't need to manually remember everything. MeMesh has **7 hooks** that capture and inject knowledge while you work:
+You don't need to manually remember everything. MeMesh has **6 hooks** that capture and inject knowledge while you work:
 
 | When | What MeMesh does |
 |------|------------------|
 | **Every session start** | Loads your most relevant memories + proactive warnings from past lessons |
 | **Before editing files** | Recalls memories tied to the file or project before Claude writes code |
-| **Before bash commands** | (Opt-in) Nudges Claude to dispatch high-verifiability commands (test, build, lint, migrate, deploy, benchmark) as background agents |
 | **When you ask to remember** | Detects "remember this" / "guardar en memesh" / "sauvegarder dans memesh" / "記下來" intent (5 languages) and reminds Claude to use memesh |
 | **After every `git commit`** | Records what you changed, with diff stats |
 | **When Claude stops** | Captures files edited, errors fixed, and auto-generates structured lessons from failures |
@@ -287,14 +307,15 @@ All configuration is via environment variables. Defaults are local-only and zero
 |---|---|---|
 | `MEMESH_DB_PATH` | `~/.memesh/knowledge-graph.db` | Override the SQLite database location. |
 | `MEMESH_AUTO_CAPTURE` | `true` | Disable the auto-capture hooks (`Stop`, `PreCompact`) entirely. |
-| `MEMESH_AUTO_DETECT_LLM` | unset (auto-detect **on**) | Set to `0` to stop memesh using an API key it finds in your shell env. By default, if `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` is set and you have not configured a provider in `~/.memesh/config.json`, memesh uses it for write-side LLM features (lesson extraction, auto-tagging, dream). Embeddings are unaffected — they stay local ONNX (384-dim) unless you explicitly set `embedder.provider`. |
-| `MEMESH_ENABLE_AGENTIC_ORCHESTRATION` | unset | Set to `1` to enable an experimental working-model protocol (CTO / Orchestrator / Agents framing). Adds a session-start banner, a Bash command nudge, and `verify_agent_work` telemetry. The protocol's effectiveness is being instrumented, not yet proven — opt in if you want to participate. **Default is OFF**: the core memory features work without this flag. |
+| `MEMESH_AUTO_DETECT_LLM` | unset (auto-detect **on**) | Set to `0` to stop memesh using an API key it finds in your shell env. By default, if `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` is set and you have not configured a provider in `~/.memesh/config.json`, memesh uses it for write-side LLM features (lesson extraction, auto-tagging, dream). Embeddings are unaffected — they stay keyword-only (FTS5) unless you explicitly set `embedder.provider` to `ollama` or `openai`. |
 | `MEMESH_AUTO_UPDATE` | `off` | Auto-update policy. `off` (default) never auto-updates; `patch` allows `X.Y.Z → X.Y.Z+N`; `minor` adds `X.Y.Z → X.Y+1.0`; `major` allows any bump. When permitted, a detached `npm install -g` fires at session end (Stop hook) so it never blocks your work — outcomes land in `~/.memesh/auto-update.log`. Also settable as `autoUpdate` in `~/.memesh/config.json` (env wins). When the installed version is deprecated by maintainers (security advisory), `patch` is force-allowed even on `off` — minor / major bumps still stay manual to avoid silent behaviour drift. |
 | `OPENAI_API_KEY` | unset | Your OpenAI key. Used automatically for LLM features unless you set `MEMESH_AUTO_DETECT_LLM=0` or configure a provider explicitly. |
 | `OLLAMA_HOST` | `http://localhost:11434` | Override the Ollama endpoint when using a local Ollama provider. |
 
 `memesh doctor` prints the resolved configuration so you can see what's active.
 
+**Fallback LLM providers (Smart Mode).** In the dashboard **Settings → "Fallback providers"** you can set an ordered failover chain — memesh tries each provider in turn when your primary is down. Add a local [Ollama](https://ollama.com) fallback, or a cloud one (OpenAI / Anthropic, with an API key). Privacy tradeoff: when a cloud fallback is used, memory text — which can be private — is sent to that provider, so it matters if you run local-only for privacy.
+
 When npm flags an installed version as deprecated (typically a security advisory), the next session-start prepends a strong `⚠️ MeMesh <ver> is DEPRECATED` banner and `memesh update-status` surfaces the same line until you upgrade. The check is cached at `~/.memesh/update-check.<version>.json` so a transient network failure can't dim the warning.
 
 ---
@@ -363,16 +384,21 @@ Or use the dashboard Settings tab (visual setup):
 memesh serve  # opens dashboard → Settings tab
 ```
 
-### Bring-your-own embeddings (optional)
+**Mine your past sessions into memory.** `memesh dream run --from-transcripts` reads this project's Claude Code session transcripts, asks the LLM for the decisions and lessons buried in the conversation, and stages them as proposals — nothing enters your graph automatically. Review each with `memesh dream show <id>` and accept the ones worth keeping. To run it on a schedule, enable `memesh config set transcriptMining true` and point a cron/launchd entry at `memesh dream run --from-transcripts --if-due` — it self-throttles (default once every 24h per project) and stays staging-only. See [API_REFERENCE](docs/api/API_REFERENCE.md#memesh-dream).
+
+### Semantic search / embeddings (optional)
 
-Embeddings default to a local ONNX model (`Xenova/all-MiniLM-L6-v2`, 384-dim) — no API key, nothing leaves your machine, and the default FTS5 recall path doesn't need them at all. To use a hosted or local-server embedder instead:
+By default MeMesh does **keyword-only** recall (FTS5) — no API key, no model download, nothing leaves your machine. Semantic (meaning-based) search is opt-in and needs an embedder. Point one of these at it:
 
 ```bash
-memesh config set embedder.provider openai          # or: ollama
+memesh config set embedder.provider ollama          # local, needs `ollama serve`
+memesh config set embedder.model nomic-embed-text
+# or, for a hosted embedder:
+memesh config set embedder.provider openai
 memesh config set embedder.model text-embedding-3-small
 ```
 
-The embedder is configured **independently of the chat LLM** — changing `llm.provider` never silently changes your embeddings. If you switch to an embedder with a different dimension (e.g. 384 → 1536), MeMesh rebuilds the vector index automatically on the next write. Supported `embedder.provider` values: `onnx` (default, local), `openai`, `ollama`.
+The embedder is configured **independently of the chat LLM** — changing `llm.provider` never silently changes your embeddings. If you switch to an embedder with a different dimension (e.g. 768 → 1536), MeMesh rebuilds the vector index automatically on the next write. Supported `embedder.provider` values: `ollama` (local), `openai` (hosted). With none set, recall stays on keyword search.
 
 | | Level 0 (default) | Level 1 (Smart Mode) |
 |---|---|---|
@@ -385,7 +411,7 @@ The embedder is configured **independently of the chat LLM** — changing `llm.p
 
 ---
 
-## All 9 Memory Tools
+## All 7 Memory Tools
 
 | Tool | What it does |
 |------|-------------|
@@ -396,7 +422,6 @@ The embedder is configured **independently of the chat LLM** — changing `llm.p
 | `import` | Import memories with merge strategies (skip / overwrite / append) |
 | `learn` | Record structured lessons from mistakes (error, root cause, fix, prevention) |
 | `user_patterns` | Analyze your work patterns — schedule, tools, strengths, learning areas |
-| `verify_agent_work` | Persist a verification report for background-agent work; reality-checks claimed file changes against `git diff` |
 
 ---
 
@@ -405,7 +430,7 @@ The embedder is configured **independently of the chat LLM** — changing `llm.p
 ```
                     ┌─────────────────┐
                     │   Core Engine   │
-                    │  (8 operations) │
+                    │  (7 operations) │
                     └────────┬────────┘
            ┌─────────────────┼─────────────────┐
            │                 │                 │

package/README.pt.md

@@ -108,7 +108,20 @@ Se você só usa memesh pelo chat do Claude Code (nunca digita `memesh` num term
 
 ## Comece em 60 Segundos
 
-### Passo 1: Instale
+### Opção A — Plugin do Claude Code (instalação em uma linha)
+
+Se você usa o Claude Code, instale o MeMesh como plugin de dentro da CLI:
+
+```
+/plugin marketplace add PCIRCLE-AI/memesh-llm-memory
+/plugin install memesh@pcircle-memesh
+```
+
+O Claude Code conecta hooks, skills e o servidor MCP automaticamente. Você ganha auto-captura em sessão, recall proativo, o skill `/memesh` na conversa e `remember` / `recall` / `forget` / `learn` como ferramentas MCP para o agente.
+
+### Opção B — npm global (otimização opcional)
+
+Se quiser o binário direto no seu `PATH` (para que `memesh` funcione em qualquer terminal sem o atraso do `npx`), ou expor `memesh-mcp` como comando stdio de caminho fixo para clientes MCP fora do Claude Code (Cursor, Cline):
 
 ```bash
 npm install -g @pcircle/memesh
@@ -127,6 +140,12 @@ Os hooks coexistem com qualquer hook customizado em `~/.claude/hooks/` — `inst
 
 ### Passo 2: Armazene uma decisão
 
+```bash
+memesh remember "Use OAuth 2.0 with PKCE for the new auth"
+```
+
+Ou use a forma explícita quando quiser um nome e um tipo estáveis para filtrar depois:
+
 ```bash
 memesh remember --name "auth-decision" --type "decision" --obs "Use OAuth 2.0 with PKCE"
 ```
@@ -234,13 +253,12 @@ Cole as ferramentas em qualquer chamada de API
 
 ## O Que Acontece Automaticamente no Claude Code
 
-Você não precisa lembrar tudo manualmente. MeMesh tem **7 hooks** que capturam e injetam conhecimento enquanto você trabalha:
+Você não precisa lembrar tudo manualmente. MeMesh tem **6 hooks** que capturam e injetam conhecimento enquanto você trabalha:
 
 | Quando | O que MeMesh faz |
 |------|------------------|
 | **Início de cada sessão** | Carrega suas memórias mais relevantes + alertas proativos de lições passadas + banner de orquestração de agentes |
 | **Antes de editar arquivos** | Recupera memórias vinculadas ao arquivo ou projeto antes de Claude escrever código |
-| **Antes de comandos bash** | Incentiva Claude a despachar comandos de alta verificabilidade (test, build, lint, migrate, deploy, benchmark) como agentes de background |
 | **Quando você pede para lembrar** | Detecta intenção "remember this" / "記下來" e lembra Claude de escrever dual (memesh + MEMORY.md) |
 | **Depois de cada `git commit`** | Registra o que você mudou, com estatísticas de diff |
 | **Quando Claude para** | Captura arquivos editados, erros corrigidos e gera automaticamente lições estruturadas de falhas |
@@ -258,14 +276,15 @@ Toda a configuração é feita por variáveis de ambiente. Os padrões são loca
 |---|---|---|
 | `MEMESH_DB_PATH` | `~/.memesh/knowledge-graph.db` | Sobrescreve a localização do banco SQLite. |
 | `MEMESH_AUTO_CAPTURE` | `true` | Desativa completamente os hooks de auto-captura (`Stop`, `PreCompact`). |
-| `MEMESH_AUTO_DETECT_LLM` | não definido (autodetecção **ligada**) | Defina como `0` para que o memesh NÃO use uma chave de API encontrada no ambiente do shell. Por padrão, se `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` estiver definida e você não tiver configurado um provedor em `~/.memesh/config.json`, o memesh a usa para as funções LLM de escrita (consolidação, extração de lições, autotagging, dream). Os embeddings não são afetados — permanecem em ONNX local (384-dim) a menos que você defina `embedder.provider` explicitamente. |
-| `MEMESH_ENABLE_AGENTIC_ORCHESTRATION` | unset | Defina como `1` para habilitar um protocolo experimental de modelo de trabalho (enquadramento CTO / Orchestrator / Agents). Adiciona um banner de início de sessão, um nudge para comandos Bash e telemetria `verify_agent_work`. A eficácia do protocolo está sendo instrumentada, ainda não comprovada — opte se quiser participar. **Padrão é OFF**: as funcionalidades de memória core funcionam sem essa flag. |
+| `MEMESH_AUTO_DETECT_LLM` | não definido (autodetecção **ligada**) | Defina como `0` para que o memesh NÃO use uma chave de API encontrada no ambiente do shell. Por padrão, se `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` estiver definida e você não tiver configurado um provedor em `~/.memesh/config.json`, o memesh a usa para as funções LLM de escrita (consolidação, extração de lições, autotagging, dream). Os embeddings não são afetados — permanecem apenas por palavras-chave (FTS5) a menos que você defina `embedder.provider` como `ollama` ou `openai`. |
 | `MEMESH_AUTO_UPDATE` | `off` | Política de auto-update. `off` (padrão) nunca faz auto-update; `patch` permite `X.Y.Z → X.Y.Z+N`; `minor` adiciona `X.Y.Z → X.Y+1.0`; `major` permite qualquer bump. Quando permitido, um `npm install -g` desanexado dispara no fim da sessão (hook Stop) para nunca bloquear seu trabalho — os resultados aparecem em `~/.memesh/auto-update.log`. Também configurável como `autoUpdate` em `~/.memesh/config.json` (env vence). Quando a versão instalada é depreciada pelos mantenedores (advisory de segurança), `patch` é forçado mesmo em `off` — bumps minor / major continuam manuais para evitar drift silencioso de comportamento. |
 | `OPENAI_API_KEY` | não definido | Sua chave da OpenAI. Usada automaticamente para as funções LLM a menos que você defina `MEMESH_AUTO_DETECT_LLM=0` ou configure um provedor explicitamente. |
 | `OLLAMA_HOST` | `http://localhost:11434` | Sobrescreve o endpoint do Ollama ao usar um provedor Ollama local. |
 
 `memesh doctor` imprime a configuração resolvida para você ver o que está ativo.
 
+**Provedores LLM de fallback (Smart Mode).** No dashboard, em **Settings → “Fallback providers”**, você pode definir uma cadeia de failover ordenada — o memesh tenta cada provedor por vez quando o principal está fora do ar. Adicione um fallback local [Ollama](https://ollama.com), ou um na nuvem (OpenAI / Anthropic, com uma API key). Compromisso de privacidade: quando um fallback na nuvem é usado, o texto da memória — que pode ser privado — é enviado a esse provedor, o que importa se você roda só local por privacidade.
+
 Quando o npm sinaliza uma versão instalada como depreciada (tipicamente um advisory de segurança), o próximo início de sessão antepõe um banner forte `⚠️ MeMesh <ver> is DEPRECATED` e `memesh update-status` mostra a mesma linha até você atualizar. A verificação fica em cache em `~/.memesh/update-check.<version>.json` para que uma falha de rede transitória não atenue o aviso.
 
 ---
@@ -334,16 +353,18 @@ Ou use a aba Settings do dashboard (setup visual):
 memesh serve  # abre dashboard → aba Settings
 ```
 
+**Minere memória das suas sessões passadas.** `memesh dream run --from-transcripts` lê as transcrições de sessão do Claude Code deste projeto, pede ao LLM as decisões e lições escondidas na conversa e as prepara como propostas — nada entra no seu grafo automaticamente. Revise cada uma com `memesh dream show <id>` e aceite as que valerem a pena.
+
 ### Use seus próprios embeddings (opcional)
 
-Os embeddings usam por padrão um modelo ONNX local (`Xenova/all-MiniLM-L6-v2`, 384-dim) — sem chave de API, nada sai da sua máquina, e o recall FTS5 padrão nem precisa deles. Para usar um embedder hospedado ou de servidor local:
+Por padrão o MeMesh faz recall **apenas por palavras-chave** (FTS5) — sem chave de API, sem download de modelo, nada sai da sua máquina. A busca semântica (por significado) é opcional e precisa de um embedder. Configure um:
 
 ```bash
 memesh config set embedder.provider openai          # or: ollama
 memesh config set embedder.model text-embedding-3-small
 ```
 
-O embedder é configurado **independentemente do LLM de chat** — mudar `llm.provider` nunca muda seus embeddings silenciosamente. Se você trocar para uma dimensão diferente (ex.: 384 → 1536), o MeMesh reconstrói o índice vetorial automaticamente na próxima escrita. Valores de `embedder.provider` suportados: `onnx` (padrão, local), `openai`, `ollama`.
+O embedder é configurado **independentemente do LLM de chat** — mudar `llm.provider` nunca muda seus embeddings silenciosamente. Se você trocar para uma dimensão diferente (ex.: 768 → 1536), o MeMesh reconstrói o índice vetorial automaticamente na próxima escrita. Valores de `embedder.provider` suportados: `ollama` (local), `openai` (hospedado). Sem nenhum, o recall permanece na busca por palavras-chave.
 
 | | Level 0 (padrão) | Level 1 (Smart Mode) |
 |---|---|---|
@@ -356,7 +377,7 @@ O embedder é configurado **independentemente do LLM de chat** — mudar `llm.pr
 
 ---
 
-## Todas as 9 Ferramentas de Memória
+## Todas as 7 Ferramentas de Memória
 
 | Ferramenta | O que faz |
 |------|-------------|
@@ -367,7 +388,6 @@ O embedder é configurado **independentemente do LLM de chat** — mudar `llm.pr
 | `import` | Importa memórias com estratégias de merge (skip / overwrite / append) |
 | `learn` | Registra lições estruturadas de erros (erro, causa raiz, fix, prevenção) |
 | `user_patterns` | Analisa seus padrões de trabalho — schedule, ferramentas, pontos fortes, áreas de aprendizado |
-| `verify_agent_work` | Persiste um relatório de verificação para trabalho de background-agent; reality-checks mudanças de arquivo declaradas contra `git diff` |
 
 ---
 
@@ -376,7 +396,7 @@ O embedder é configurado **independentemente do LLM de chat** — mudar `llm.pr
 ```
                     ┌─────────────────┐
                     │   Core Engine   │
-                    │  (8 operations) │
+                    │  (7 operations) │
                     └────────┬────────┘
            ┌─────────────────┼─────────────────┐
            │                 │                 │

package/README.th.md

@@ -142,6 +142,12 @@ Hooks เหล่านี้อยู่ร่วมกับ custom hooks ท
 
 ### ขั้นตอนที่ 2: เก็บการตัดสินใจ
 
+```bash
+memesh remember "Use OAuth 2.0 with PKCE for the new auth"
+```
+
+หรือใช้รูปแบบชัดเจนเมื่อต้องการชื่อและชนิดที่คงที่สำหรับกรองภายหลัง:
+
 ```bash
 memesh remember --name "auth-decision" --type "decision" --obs "Use OAuth 2.0 with PKCE"
 ```
@@ -249,13 +255,12 @@ memesh export-schema \
 
 ## สิ่งที่เกิดขึ้นโดยอัตโนมัติใน Claude Code
 
-ไม่ต้องจำข้อมูลทุกอย่างด้วยตนเอง MeMesh มี **7 hook** ที่บันทึกและแทรกความรู้ขณะทำงาน:
+ไม่ต้องจำข้อมูลทุกอย่างด้วยตนเอง MeMesh มี **6 hook** ที่บันทึกและแทรกความรู้ขณะทำงาน:
 
 | เมื่อ | MeMesh ทำอะไร |
 |---|---|
 | **เริ่มเซสชันทุกครั้ง** | โหลดหน่วยความจำที่เกี่ยวข้องมากที่สุด + คำเตือนจากบทเรียนที่ผ่านมา + แบนเนอร์การ orchestrate เอเจนต์ |
 | **ก่อนแก้ไขไฟล์** | เรียกคืนหน่วยความจำที่เชื่อมโยงกับไฟล์หรือโครงการก่อนที่ Claude เขียนโค้ด |
-| **ก่อนคำสั่ง bash** | แนะนำให้ Claude สั่งคำสั่งที่ยืนยันได้สูง (test build lint migrate deploy benchmark) เป็นเอเจนต์พื้นหลัง |
 | **เมื่อคุณขอให้จำ** | ตรวจจับความตั้งใจ "remember this" / "記下來" และเตือน Claude ให้เขียนแบบสองทาง (memesh + MEMORY.md) |
 | **หลังทุก `git commit`** | บันทึกสิ่งที่เปลี่ยนแปลง พร้อมสถิติ diff |
 | **เมื่อ Claude หยุด** | บันทึกไฟล์ที่แก้ไข บั๊กที่แก้ไข และสร้างบทเรียนโครงสร้างจากความล้มเหลวโดยอัตโนมัติ |
@@ -273,14 +278,15 @@ memesh export-schema \
 |---|---|---|
 | `MEMESH_DB_PATH` | `~/.memesh/knowledge-graph.db` | เปลี่ยนตำแหน่งฐานข้อมูล SQLite |
 | `MEMESH_AUTO_CAPTURE` | `true` | ปิดการใช้ hook จับข้อมูลอัตโนมัติทั้งหมด (`Stop`, `PreCompact`) |
-| `MEMESH_AUTO_DETECT_LLM` | ไม่ได้ตั้งค่า (ตรวจจับอัตโนมัติ **เปิด**) | ตั้งเป็น `0` เพื่อไม่ให้ memesh ใช้คีย์ API ที่พบในสภาพแวดล้อมของเชลล์ โดยค่าเริ่มต้น หากตั้ง `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` ไว้ และคุณยังไม่ได้กำหนดผู้ให้บริการใน `~/.memesh/config.json` memesh จะใช้คีย์นั้นสำหรับฟีเจอร์ LLM ฝั่งเขียน (การสกัดบทเรียน, auto-tagging, dream) ส่วน embeddings ไม่ได้รับผลกระทบ — ยังคงเป็น ONNX ในเครื่อง (384 มิติ) เว้นแต่คุณจะตั้ง `embedder.provider` อย่างชัดเจน |
-| `MEMESH_ENABLE_AGENTIC_ORCHESTRATION` | ไม่ตั้ง | ตั้งเป็น `1` เพื่อเปิดใช้โปรโตคอล working-model เชิงทดลอง (กรอบ CTO / Orchestrator / Agents) เพิ่มแบนเนอร์ตอนเริ่มเซสชัน การเตือนคำสั่ง Bash และเทเลเมตรี `verify_agent_work` ประสิทธิผลของโปรโตคอลกำลังถูกเก็บข้อมูล ยังไม่ได้พิสูจน์ — opt-in ถ้าต้องการเข้าร่วม **ค่าเริ่มต้นปิด**: ฟีเจอร์หน่วยความจำหลักทำงานได้โดยไม่ต้องเปิดธงนี้ |
+| `MEMESH_AUTO_DETECT_LLM` | ไม่ได้ตั้งค่า (ตรวจจับอัตโนมัติ **เปิด**) | ตั้งเป็น `0` เพื่อไม่ให้ memesh ใช้คีย์ API ที่พบในสภาพแวดล้อมของเชลล์ โดยค่าเริ่มต้น หากตั้ง `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OLLAMA_HOST` ไว้ และคุณยังไม่ได้กำหนดผู้ให้บริการใน `~/.memesh/config.json` memesh จะใช้คีย์นั้นสำหรับฟีเจอร์ LLM ฝั่งเขียน (การสกัดบทเรียน, auto-tagging, dream) ส่วน embeddings ไม่ได้รับผลกระทบ — ยังคงเป็นการค้นหาด้วยคีย์เวิร์ดอย่างเดียว (FTS5) เว้นแต่คุณจะตั้ง `embedder.provider` เป็น `ollama` หรือ `openai` |
 | `MEMESH_AUTO_UPDATE` | `off` | นโยบายอัปเดตอัตโนมัติ `off` (ค่าเริ่มต้น) ไม่อัปเดตเลย; `patch` อนุญาต `X.Y.Z → X.Y.Z+N`; `minor` เพิ่ม `X.Y.Z → X.Y+1.0`; `major` อนุญาตทุกการเพิ่มเวอร์ชัน เมื่ออนุญาต `npm install -g` แบบ detached จะทำงานเมื่อจบเซสชัน (Stop hook) เพื่อไม่บล็อกงานของคุณ — ผลลัพธ์ลงใน `~/.memesh/auto-update.log` ตั้งใน `~/.memesh/config.json` ผ่านคีย์ `autoUpdate` ก็ได้ (env ชนะ) เมื่อเวอร์ชันที่ติดตั้งถูก deprecate (security advisory) `patch` จะถูกบังคับเปิดแม้ตั้งเป็น `off` — minor / major ยังต้องทำมือเพื่อหลีกเลี่ยงการเปลี่ยนพฤติกรรมเงียบ ๆ |
 | `OPENAI_API_KEY` | ไม่ได้ตั้งค่า | คีย์ OpenAI ของคุณ ใช้โดยอัตโนมัติสำหรับฟีเจอร์ LLM เว้นแต่คุณจะตั้ง `MEMESH_AUTO_DETECT_LLM=0` หรือกำหนดผู้ให้บริการอย่างชัดเจน |
 | `OLLAMA_HOST` | `http://localhost:11434` | เปลี่ยนปลายทาง Ollama เมื่อใช้ผู้ให้บริการ Ollama ภายในเครื่อง |
 
 `memesh doctor` พิมพ์การตั้งค่าที่ resolve แล้วเพื่อให้คุณเห็นว่าอะไรทำงานอยู่
 
+**ผู้ให้บริการ LLM สำรอง (Smart Mode)** ใน **Settings → “Fallback providers”** ของ dashboard คุณตั้งลูกโซ่ failover ตามลำดับได้ — เมื่อผู้ให้บริการหลักล่ม memesh จะลองตัวถัดไปในรายการตามลำดับ เพิ่มตัวสำรองแบบโลคัล [Ollama](https://ollama.com) หรือแบบคลาวด์ (OpenAI / Anthropic ต้องมี API key) ก็ได้ ข้อแลกเปลี่ยนด้านความเป็นส่วนตัว: เมื่อใช้ตัวสำรองแบบคลาวด์ ข้อความในหน่วยความจำ (ซึ่งอาจเป็นข้อมูลส่วนตัว) จะถูกส่งไปยังผู้ให้บริการนั้น จึงสำคัญถ้าคุณรันแบบโลคัลล้วนเพื่อความเป็นส่วนตัว
+
 เมื่อ npm ระบุว่าเวอร์ชันที่ติดตั้งถูก deprecate (โดยทั่วไปคือ security advisory) เซสชันถัดไปจะแสดงแบนเนอร์ `⚠️ MeMesh <ver> is DEPRECATED` แบบหนักนำหน้า และ `memesh update-status` จะแสดงบรรทัดเดียวกันจนกว่าคุณจะอัปเกรด การตรวจสอบถูก cache ที่ `~/.memesh/update-check.<version>.json` เพื่อไม่ให้ความล้มเหลวเครือข่ายชั่วคราวลดความสว่างของคำเตือน
 
 ---
@@ -349,16 +355,18 @@ memesh config set llm.api-key sk-ant-...
 memesh serve  # opens dashboard → Settings tab
 ```
 
+**ขุดเซสชันที่ผ่านมาให้เป็นหน่วยความจำ** `memesh dream run --from-transcripts` จะอ่านบันทึกเซสชัน Claude Code ของโปรเจกต์นี้ ถาม LLM หาการตัดสินใจและบทเรียนที่ซ่อนอยู่ในบทสนทนา แล้วพักไว้เป็นข้อเสนอ — ไม่มีอะไรเข้าสู่กราฟความรู้โดยอัตโนมัติ ตรวจดูทีละรายการด้วย `memesh dream show <id>` แล้ว accept เฉพาะอันที่ควรเก็บ
+
 ### ใช้ embeddings ของคุณเอง (ไม่บังคับ)
 
-โดยค่าเริ่มต้น embeddings ใช้โมเดล ONNX ในเครื่อง (`Xenova/all-MiniLM-L6-v2`, 384 มิติ) — ไม่ต้องใช้คีย์ API ไม่มีข้อมูลออกจากเครื่อง และการ recall แบบ FTS5 เริ่มต้นก็ไม่ต้องใช้เลย หากต้องการใช้ embedder แบบโฮสต์หรือเซิร์ฟเวอร์ในเครื่อง:
+โดยค่าเริ่มต้น MeMesh ทำ recall **ด้วยคีย์เวิร์ดอย่างเดียว** (FTS5) — ไม่ต้องใช้คีย์ API ไม่ต้องดาวน์โหลดโมเดล ไม่มีข้อมูลออกจากเครื่อง การค้นหาเชิงความหมาย (semantic) เป็นตัวเลือกเสริมและต้องใช้ embedder ตั้งค่าอย่างใดอย่างหนึ่ง:
 
 ```bash
 memesh config set embedder.provider openai          # or: ollama
 memesh config set embedder.model text-embedding-3-small
 ```
 
-embedder ถูกตั้งค่า**แยกจาก LLM แชท** — การเปลี่ยน `llm.provider` จะไม่เปลี่ยน embeddings ของคุณอย่างเงียบ ๆ หากเปลี่ยนไปใช้มิติที่ต่างกัน (เช่น 384 → 1536) MeMesh จะสร้างดัชนีเวกเตอร์ใหม่โดยอัตโนมัติในการเขียนครั้งถัดไป ค่า `embedder.provider` ที่รองรับ: `onnx` (ค่าเริ่มต้น ในเครื่อง), `openai`, `ollama`
+embedder ถูกตั้งค่า**แยกจาก LLM แชท** — การเปลี่ยน `llm.provider` จะไม่เปลี่ยน embeddings ของคุณอย่างเงียบ ๆ หากเปลี่ยนไปใช้มิติที่ต่างกัน (เช่น 768 → 1536) MeMesh จะสร้างดัชนีเวกเตอร์ใหม่โดยอัตโนมัติในการเขียนครั้งถัดไป ค่า `embedder.provider` ที่รองรับ: `ollama` (ในเครื่อง), `openai` (โฮสต์) หากไม่ตั้งค่าใดเลย recall จะยังคงเป็นการค้นหาด้วยคีย์เวิร์ด
 
 | | ระดับ 0 (ค่าเริ่มต้น) | ระดับ 1 (Smart Mode) |
 |---|---|---|
@@ -371,7 +379,7 @@ embedder ถูกตั้งค่า**แยกจาก LLM แชท** —
 
 ---
 
-## เครื่องมือหน่วยความจำทั้ง 9 ตัว
+## เครื่องมือหน่วยความจำทั้ง 7 ตัว
 
 | เครื่องมือ | ทำอะไร |
 |---|---|
@@ -382,7 +390,6 @@ embedder ถูกตั้งค่า**แยกจาก LLM แชท** —
 | `import` | นำเข้าหน่วยความจำพร้อมกลยุทธ์ผสาน (ข้าม / เขียนทับ / ผนวก) |
 | `learn` | บันทึกบทเรียนโครงสร้างจากข้อผิดพลาด (ข้อผิดพลาด สาเหตุ การแก้ไข การป้องกัน) |
 | `user_patterns` | วิเคราะห์รูปแบบการทำงาน — ตารางเวลา เครื่องมือ จุดแข็ง พื้นที่เรียนรู้ |
-| `verify_agent_work` | คงรายงานการยืนยันสำหรับงานเอเจนต์พื้นหลัง ตรวจสอบความเป็นจริงการเปลี่ยนแปลงไฟล์ที่อ้างสิทธิ์เทียบกับ `git diff` |
 
 ---
 
@@ -391,7 +398,7 @@ embedder ถูกตั้งค่า**แยกจาก LLM แชท** —
 ```
                     ┌─────────────────┐
                     │   Core Engine   │
-                    │  (8 operations) │
+                    │  (7 operations) │
                     └────────┬────────┘
            ┌─────────────────┼─────────────────┐
            │                 │                 │