memorylake-openclaw 0.0.3 → 0.0.5-beta.1

package/.claude/settings.local.json

@@ -0,0 +1,22 @@
+{
+  "permissions": {
+    "allow": [
+      "WebFetch(domain:10.71.10.71)",
+      "Bash(find /Users/henry/work/github/memorylake-openclaw -maxdepth 2 \\\\\\( -name \".env*\" -o -name \"*.config.*\" -o -name \"openclaw.json\" \\\\\\) 2>/dev/null | grep -v node_modules)",
+      "Bash(jq '.env' /Users/henry/.openclaw/openclaw.json)",
+      "Bash(wc -l /Users/henry/.openclaw/agents/main/sessions/*.jsonl)",
+      "Bash(curl -s http://10.71.10.71:8001/v3/api-docs/open-api 2>&1 | head -5000)",
+      "Bash(cat /Users/henry/.claude/projects/-Users-henry-work-github-memorylake-openclaw/667ef3f7-2edf-4030-b7cf-239a4d916489/tool-results/bf9jm64sc.txt | python3 -c \"\nimport json, sys\ndata = json.load\\(sys.stdin\\)\n# Extract the memories endpoint\npaths = data.get\\('paths', {}\\)\nfor path, methods in paths.items\\(\\):\n    if 'memories' in path:\n        print\\(f'\\\\\\\\n=== {path} ==='\\)\n        for method, spec in methods.items\\(\\):\n            print\\(f'\\\\\\\\n--- {method.upper\\(\\)} ---'\\)\n            print\\(json.dumps\\(spec, indent=2\\)\\)\n\n# Also print relevant schemas\nschemas = data.get\\('components', {}\\).get\\('schemas', {}\\)\nfor name, schema in schemas.items\\(\\):\n    if 'memory' in name.lower\\(\\) or 'Memory' in name or 'Message' in name:\n        print\\(f'\\\\\\\\n=== Schema: {name} ==='\\)\n        print\\(json.dumps\\(schema, indent=2\\)\\)\n\")",
+      "Bash(cat /Users/henry/.openclaw/openclaw.json | python3 -c \"\nimport json, sys\ndata = json.load\\(sys.stdin\\)\n# Look for plugin-related config\nfor key in data:\n    if 'plugin' in key.lower\\(\\) or 'memory' in key.lower\\(\\) or 'memorylake' in key.lower\\(\\) or 'extension' in key.lower\\(\\):\n        print\\(f'=== {key} ==='\\)\n        print\\(json.dumps\\(data[key], indent=2\\)[:3000]\\)\n# Also print top-level keys\nprint\\('\\\\\\\\n=== Top-level keys ==='\\)\nprint\\(list\\(data.keys\\(\\)\\)\\)\n\")",
+      "Bash(cat /Users/henry/.openclaw/agents/main/sessions/sessions.json | python3 -c \"\nimport json, sys\ndata = json.load\\(sys.stdin\\)\nfor key, val in data.items\\(\\):\n    print\\(f'Key: {key}'\\)\n    print\\(f'  sessionId: {val.get\\(\\\\\"sessionId\\\\\"\\)}'\\)\n    print\\(f'  chatType: {val.get\\(\\\\\"chatType\\\\\"\\)}'\\)\n    print\\(f'  channel: {val.get\\(\\\\\"channel\\\\\"\\)}'\\)\n    print\\(f'  displayName: {val.get\\(\\\\\"displayName\\\\\"\\)}'\\)\n    print\\(f'  subject: {val.get\\(\\\\\"subject\\\\\"\\)}'\\)\n    print\\(f'  sessionFile: {val.get\\(\\\\\"sessionFile\\\\\"\\)}'\\)\n    print\\(\\)\n\")",
+      "Bash(head -20 /Users/henry/.openclaw/agents/main/sessions/8cc2bff3-a3af-4725-b071-43caffb36771.jsonl | python3 -c \"\nimport json, sys\nfor line in sys.stdin:\n    line = line.strip\\(\\)\n    if not line:\n        continue\n    obj = json.loads\\(line\\)\n    t = obj.get\\('type'\\)\n    if t == 'message':\n        msg = obj.get\\('message', {}\\)\n        role = msg.get\\('role'\\)\n        content = msg.get\\('content', []\\)\n        # Truncate content for display\n        content_str = str\\(content\\)[:200]\n        print\\(f'Type: {t}, Role: {role}, Content: {content_str}...'\\)\n    else:\n        print\\(f'Type: {t}, Keys: {list\\(obj.keys\\(\\)\\)}'\\)\n\")",
+      "Bash(ls ~/.openclaw/agents/main/sessions/*.jsonl 2>/dev/null | head -3)",
+      "Bash(cat ~/.openclaw/agents/main/sessions/sessions.json 2>/dev/null | python3 -c \"import json,sys; d=json.load\\(sys.stdin\\); print\\(json.dumps\\(dict\\(list\\(d.items\\(\\)\\)[:2]\\), indent=2\\)\\)\" 2>/dev/null || echo \"No sessions.json or parse error\")",
+      "Bash(head -5 ~/.openclaw/agents/main/sessions/372112f5-ea26-4440-995f-aa756c707858.jsonl 2>/dev/null | python3 -c \"\nimport json, sys\nfor line in sys.stdin:\n    line = line.strip\\(\\)\n    if not line: continue\n    d = json.loads\\(line\\)\n    print\\(json.dumps\\({k: d[k] for k in list\\(d.keys\\(\\)\\)[:3]}, indent=2, ensure_ascii=False\\)\\)\n    if d.get\\('type'\\) == 'message':\n        msg = d.get\\('message', {}\\)\n        role = msg.get\\('role'\\)\n        content = msg.get\\('content'\\)\n        if isinstance\\(content, list\\):\n            types = [c.get\\('type'\\) for c in content]\n            print\\(f'  role={role}, content_types={types}'\\)\n        else:\n            print\\(f'  role={role}, content type={type\\(content\\).__name__}'\\)\n\")",
+      "Bash(python3 -c \"\nimport json\nwith open\\('/Users/henry/.openclaw/agents/main/sessions/372112f5-ea26-4440-995f-aa756c707858.jsonl'\\) as f:\n    for line in f:\n        line = line.strip\\(\\)\n        if not line: continue\n        d = json.loads\\(line\\)\n        if d.get\\('type'\\) != 'message': continue\n        msg = d.get\\('message', {}\\)\n        role = msg.get\\('role'\\)\n        content = msg.get\\('content'\\)\n        if isinstance\\(content, list\\):\n            types = [c.get\\('type'\\) for c in content]\n            print\\(f'role={role}, content_types={types}'\\)\n        elif isinstance\\(content, str\\):\n            print\\(f'role={role}, content=str[{len\\(content\\)}]'\\)\n        else:\n            print\\(f'role={role}, content={type\\(content\\).__name__}'\\)\n\" | head -20)",
+      "Bash(python3 << 'PYEOF'\nimport json\nwith open\\(\"/Users/henry/.openclaw/agents/main/sessions/372112f5-ea26-4440-995f-aa756c707858.jsonl\"\\) as f:\n    for line in f:\n        line = line.strip\\(\\)\n        if not line:\n            continue\n        d = json.loads\\(line\\)\n        if d.get\\(\"type\"\\) != \"message\":\n            continue\n        msg = d.get\\(\"message\", {}\\)\n        role = msg.get\\(\"role\"\\)\n        content = msg.get\\(\"content\"\\)\n        if isinstance\\(content, list\\):\n            types = [c.get\\(\"type\"\\) for c in content]\n            print\\(f\"role={role}, content_types={types}\"\\)\n        elif isinstance\\(content, str\\):\n            print\\(f\"role={role}, content=str[{len\\(content\\)}]\"\\)\n        else:\n            print\\(f\"role={role}, content={type\\(content\\).__name__}\"\\)\nPYEOF)",
+      "Bash(python3 << 'PYEOF'\nimport json\nwith open\\(\"/Users/henry/.openclaw/openclaw.json\"\\) as f:\n    config = json.load\\(f\\)\n# Show agents structure\nagents = config.get\\(\"agents\", {}\\)\nprint\\(json.dumps\\({\n    \"defaults\": agents.get\\(\"defaults\", {}\\),\n    \"list_keys\": [list\\(a.keys\\(\\)\\) for a in agents.get\\(\"list\", []\\)][:3],\n    \"list_sample\": agents.get\\(\"list\", []\\)[:2]\n}, indent=2, default=str\\)\\)\nPYEOF)",
+      "Bash(chmod +x /Users/henry/work/github/memorylake-openclaw/skills/migrate-memories-to-memorylake/migrate.mjs)"
+    ]
+  }
+}

package/.cursor/pr-review-2.md

@@ -0,0 +1,79 @@
+## PR #2 Review：Add advanced web search tool with plugin-level constraints
+
+> 之前那条简短英文评论格式不太好，这条为主，请以本评论为准。
+
+### 总体评价
+- **整体设计合理**：`advanced_web_search` 作为可选工具（`optional: true`），需要在 OpenClaw 侧显式允许，和现有工具体系一致。
+- **配置与文档同步到位**：新增的 `webSearch*` 配置在 README、OpenClaw 文档以及 `openclaw.plugin.json` 里都有说明，label/placeholder/help 也比较清晰。
+- **错误处理风格统一**：工具实现中的 try/catch、返回 `content` + `details` 的方式，与现有 `document_search` 等工具保持一致。
+
+### 需要在合并前确认/修复的点（Blocking）
+1. **Web Search API 路径**
+
+   ```ts
+   this.webSearchPath = "api/v1/search";  //TODO: update to the new web search API `openapi/memorylake/api/v1/search`
+   ```
+
+   - 目前这里用的是 `api/v1/search`，但其它路径（memories/doc search）都是走 `openapi/memorylake/...`。
+   - TODO 里也写了目标路径可能是 `openapi/memorylake/api/v1/search`。
+   - **建议**：跟后端或 API 文档确认最终路径，二选一：
+     - 若最终是 `openapi/memorylake/api/v1/search`，这里改成该路径并删除 TODO；
+     - 若确实是 `api/v1/search`，则保留该值、删掉 TODO，并补一行注释说明这是 web search 的独立 endpoint。
+
+2. **统一搜索 API 的响应结构**
+
+   - 目前 `searchWeb` 是：
+
+     ```ts
+     const resp = await this.http
+       .post(this.webSearchPath, { json: body })
+       .json<WebSearchResponse>();
+     return normalizeWebSearchResponse(resp);
+     ```
+
+   - 但其它接口（memories/doc search）是走统一 `ApiResponse` 包装：先检查 `success`，再从 `data` 里取结果。
+   - 如果 unified web search API 实际返回的是 `{ success, data: { results, total_results } }` 这一套，那么这里会少了解包，`normalizeWebSearchResponse` 的入参就不对齐。
+   - **建议**：确认该接口真实响应：
+     - 若也是 `ApiResponse` 风格，就改成先解析 `ApiResponse`（含 `success` 判断），再对 `data` 做 `normalizeWebSearchResponse`，与其它接口保持一致；
+     - 若这个接口是裸的 `WebSearchResponse`（没有 `success/data` 包装），当前写法就 OK，可以在注释里说明这是一个特例。
+
+### 非阻塞的改进建议（Nice to have）
+1. **空数组语义**
+
+   - 现在发送请求前的判断是：
+
+     ```ts
+     if (options.include_domains?.length) body.include_domains = options.include_domains;
+     if (options.exclude_domains?.length) body.exclude_domains = options.exclude_domains;
+     ```
+
+   - 这意味着：当配置为 `[]` 时字段不会发送，仅当长度 > 0 才会发。
+   - 如果产品上不需要区分：“未配置” vs “显式设为空数组”，当前实现是合理的；
+   - 如果未来希望支持“显式禁用所有域名”之类的语义，可能需要改成 `options.include_domains !== undefined` 决定是否传字段。
+
+2. **normalizeWebSearchResponse 的稳健性**
+
+   - 当前只对顶层 `results`/`total_results` 做了类型兜底：
+
+     ```ts
+     return {
+       results: Array.isArray(raw?.results) ? raw.results : [],
+       total_results: typeof raw?.total_results === "number" ? raw.total_results : 0,
+     };
+     ```
+
+   - 如果后端在单条 result 的字段上（`url`/`title` 等）有可能返回 `null` 或非字符串，将来可以考虑在这里顺便做一层浅 normalize，减少下游使用时的防御性代码。不是必须，看 API 稳定性。
+
+3. **配置解析的测试**
+
+   - 新增的 `parseOptionalStringArray` / `parseOptionalString` 实现本身比较简单，但关系到配置体验。
+   - 如果项目中已有 config 相关的测试套件，可以考虑加一小组用例（合法/非法类型，`null`、`[]` 等），防止以后改动时出现回归。
+
+### 结论
+
+- **总体 LGTM**：设计、类型、文档和插件配置都比较完整。
+- 真正阻塞合并的主要是：
+  - 确认并固定 `webSearchPath` 路径；
+  - 确认统一搜索 API 的响应是否包在 `ApiResponse` 里，并根据结果决定是否需要像其它接口那样做一层解包。
+- 这两个点确认/修完之后，就可以放心合并了。
+

package/.cursor/skills/gh-pr-description-gen/SKILL.md

@@ -0,0 +1,83 @@
+---
+name: gh-pr-description-gen
+description: Use when creating a GitHub PR with gh CLI and needing to generate title and description from git diff and optional issue context
+---
+
+# gh PR Create with Generated Description
+
+## Overview
+
+Create a GitHub pull request using `gh pr create`, with title and body generated by the agent from the git diff. No external model calls—the agent generates the content directly.
+
+## When to Use
+
+- User wants to create a PR with `gh` and needs title/description written
+- User has uncommitted or committed changes and wants a PR created
+- User may reference related issues (e.g. "fixes #123") to include in context
+
+## Core Workflow
+
+1. **Gather context**
+   - Current branch: `git branch --show-current`
+   - Base branch: `git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's#.*/##'` or fallback to `main`/`master`
+   - Git diff: `git diff origin/{base}...HEAD` (or `git diff origin/{base}` if no merge base)
+   - Truncate diff to ~15000 chars if large
+   - Optional: if user provides issue numbers, run `gh issue view {number}` for each to get title and description
+
+2. **Generate title and body**
+   - Use the diff (and optionally issue context) to write a clear title and description
+   - Title: concise, descriptive, no "to #X:" prefix
+   - Body format:
+
+```
+### Motivation
+
+(Describe the motivation of this PR)
+
+### Modifications
+
+* (Modification 1)
+* (Modification 2)
+* ...
+```
+
+   - Use fluent, simple-yet-elegant English
+   - Keep meanings clear, sentences short, lines under ~150 characters
+   - If related to issues, add "Fixes #123" or "Closes #123" in body to auto-link
+
+3. **Create PR**
+   - `gh pr create --title "..." --body "..."` (or `--body-file -` with stdin)
+   - Add `--base` if base branch differs from default
+   - Add `--draft` if user wants a draft PR
+
+## Quick Reference
+
+| Step | Command |
+|------|---------|
+| Current branch | `git branch --show-current` |
+| Default base | `git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null \| sed 's#.*/##'` or `main` |
+| Diff | `git diff origin/{base}...HEAD` |
+| Issue view | `gh issue view {number} --json title,body` |
+| Create PR | `gh pr create -t "Title" -b "Body"` |
+
+## Body Template
+
+Use this structure when generating the description:
+
+```markdown
+### Motivation
+
+(Describe what problem this PR solves or why it matters)
+
+### Modifications
+
+* (First change)
+* (Second change)
+* ...
+```
+
+## Common Mistakes
+
+- **Forgetting to push**: Ensure branch is pushed before `gh pr create`; `gh` will prompt if not
+- **Long body**: Use `--body-file -` with heredoc when body is large or contains special chars
+- **Wrong base**: Use `--base branch` when base differs from default

package/.cursor/skills/git-create-branch/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: git-create-branch
+description: 标准化流程：先从远端获取最新 main，再创建 feature/bugfix 分支用于功能开发或 bug 修复。当你准备开始一项新开发任务并需要新分支时使用。
+---
+
+# Git 新建分支流程（基于 main）
+
+## 前置约定
+
+- **默认远端名**：`origin`（如果项目不是用 `origin`，请将下面命令中的 `origin` 换成实际远端名）
+- **主干分支名**：`main`（如果项目主干是 `master` 或其他名字，同理替换）
+- **前提条件**：在运行本流程前，工作区应当是干净的（没有未提交更改，或明确知道自己在做什么）
+
+## 一、检查当前工作区状态
+
+1. 查看当前分支与工作区：
+
+   ```bash
+   git status
+   ```
+
+2. 如果有未提交的更改：
+   - 需要的话先提交：`git commit -am "your message"`
+   - 或者暂存：`git stash push -m "temp before new branch"`
+   - 或者放弃本地修改：`git restore .`（谨慎使用，会丢弃修改）
+
+目标：在继续之前，`git status` 应尽量是干净状态。
+
+## 二、从远端获取最新 main
+
+1. 抓取远端最新记录（包括 main）：
+
+   ```bash
+   git fetch origin main
+   ```
+
+2. 切换到本地 `main` 分支（如果当前还不在）：
+
+   ```bash
+   git checkout main
+   ```
+
+3. 将本地 `main` 更新到与远端一致，避免产生多余 merge 提交：
+
+   ```bash
+   git pull --ff-only origin main
+   ```
+
+如果 `--ff-only` 报错，说明本地 `main` 有额外提交，需要先确认是否应该保留；通常建议保持本地 `main` 与远端完全一致，没有额外提交。
+
+## 三、从最新 main 创建开发分支
+
+1. 根据需求选择分支前缀：
+   - 功能开发：`feature/xxx-简短描述`
+   - Bug 修复：`bugfix/xxx-简短描述`
+
+   示例：
+   - `feature/1234-add-knowledge-search`
+   - `bugfix/5678-fix-memory-sync`
+
+2. 从最新 `main` 创建并切换到新分支（将 `<new-branch-name>` 替换为你的实际分支名）：
+
+   ```bash
+   git checkout -b <new-branch-name> main
+   ```
+
+3. 确认当前分支确实是新建分支：
+
+   ```bash
+   git branch --show-current
+   git status
+   ```
+
+## 四、（可选）立即将新分支推送到远端
+
+如果你希望尽早在远端创建对应分支（方便备份或协作）：
+
+```bash
+git push -u origin <new-branch-name>
+```
+
+- `-u` 会设置默认上游分支，之后只需要 `git push` / `git pull` 即可。
+
+## 五、后续使用建议
+
+- **开始开发**：在新分支上进行所有与本任务相关的修改与提交。
+- **保持同步 main**：在开发周期较长时，周期性地从 `main` 合并或 rebase，避免与主干差异过大，例如：
+
+  ```bash
+  git checkout main
+  git pull --ff-only origin main
+  git checkout <new-branch-name>
+  git rebase main   # 或 git merge main，根据团队习惯
+  ```
+
+- **完成开发后**：通过 PR / Merge Request 将该分支合入主干，然后根据团队流程删除本地与远端分支。
+

package/README.md

@@ -10,7 +10,7 @@ Your agent forgets everything between sessions. This plugin fixes that. It watch
   <img src="../docs/images/openclaw-architecture.png" alt="Architecture" width="800" />
 </p> -->
 
-**Auto-Recall** — Before the agent responds, the plugin searches MemoryLake for memories that match the current message and injects them into context.
+**Auto-Recall** — Before the agent responds, the plugin searches MemoryLake for memories and relevant document excerpts that match the current message and injects them into context.
 
 **Auto-Capture** — After the agent responds, the plugin sends the exchange to MemoryLake. MemoryLake decides what's worth keeping — new facts get stored, stale ones updated, duplicates merged.
 
@@ -29,16 +29,15 @@ Get an API key from [app.memorylake.ai](https://app.memorylake.ai), then add to
 "memorylake-openclaw": {
   "enabled": true,
   "config": {
-    "apiKey": "${MEMORYLAKE_API_KEY}",
-    "projectId": "proj-...",
-    "userId": "your-user-id"
+    "apiKey": "sk-...",
+    "projectId": "proj-..."
   }
 }
 ```
 
 ## Agent tools
 
-The agent gets five tools it can call during conversations:
+The agent gets six tools it can call during conversations:
 
 | Tool | Description |
 |------|-------------|
@@ -47,6 +46,7 @@ The agent gets five tools it can call during conversations:
 | `memory_store` | Explicitly save a fact |
 | `memory_get` | Retrieve a memory by ID |
 | `memory_forget` | Delete a memory by ID |
+| `document_search` | Search project documents for relevant paragraphs, tables, and figures |
 
 ## CLI
 
@@ -65,7 +65,6 @@ openclaw memorylake stats
 | `apiKey` | `string` | — | **Required.** MemoryLake API key (supports `${MEMORYLAKE_API_KEY}`) |
 | `projectId` | `string` | — | **Required.** MemoryLake project ID |
 | `host` | `string` | `https://app.memorylake.ai` | MemoryLake server endpoint URL |
-| `userId` | `string` | `"default"` | Scope memories per user |
 | `autoRecall` | `boolean` | `true` | Inject memories before each turn |
 | `autoCapture` | `boolean` | `true` | Store facts after each turn |
 | `topK` | `number` | `5` | Max memories per recall |

package/docs/openclaw.mdx

@@ -11,9 +11,9 @@ Add long-term memory to [OpenClaw](https://github.com/openclaw/openclaw) agents
 </Frame>*/}
 
 The plugin provides:
-1. **Auto-Recall** — Before the agent responds, memories matching the current message are injected into context
+1. **Auto-Recall** — Before the agent responds, memories and relevant document excerpts matching the current message are injected into context
 2. **Auto-Capture** — After the agent responds, the exchange is sent to MemoryLake which decides what's worth keeping
-3. **Agent Tools** — Five tools for explicit memory operations during conversations
+3. **Agent Tools** — Six tools for memory and document operations during conversations
 
 Both auto-recall and auto-capture run silently with no manual configuration required.
 
@@ -34,16 +34,15 @@ Add to your `openclaw.json`:
 "memorylake-openclaw": {
   "enabled": true,
   "config": {
-    "apiKey": "${MEMORYLAKE_API_KEY}",
-    "projectId": "proj-...",
-    "userId": "your-user-id"
+    "apiKey": "sk-...",
+    "projectId": "proj-..."
   }
 }
 ```
 
 ## Agent Tools
 
-The agent gets five tools it can call during conversations:
+The agent gets six tools it can call during conversations:
 
 | Tool | Description |
 |------|-------------|
@@ -52,6 +51,7 @@ The agent gets five tools it can call during conversations:
 | `memory_store` | Explicitly save a fact |
 | `memory_get` | Retrieve a memory by ID |
 | `memory_forget` | Delete a memory by ID |
+| `document_search` | Search project documents for relevant paragraphs, tables, and figures |
 
 ## CLI Commands
 
@@ -70,7 +70,6 @@ openclaw memorylake stats
 | `apiKey` | `string` | — | **Required.** MemoryLake API key (supports `${MEMORYLAKE_API_KEY}`) |
 | `projectId` | `string` | — | **Required.** MemoryLake project ID |
 | `host` | `string` | `https://app.memorylake.ai` | MemoryLake server endpoint URL |
-| `userId` | `string` | `"default"` | Scope memories per user |
 | `autoRecall` | `boolean` | `true` | Inject memories before each turn |
 | `autoCapture` | `boolean` | `true` | Store facts after each turn |
 | `topK` | `number` | `5` | Max memories per recall |
@@ -82,7 +81,7 @@ openclaw memorylake stats
 1. **Zero Configuration** — Auto-recall and auto-capture work out of the box with no prompting required
 2. **Async Processing** — Memory extraction runs asynchronously via MemoryLake's API
 3. **Session Tracking** — Conversations are tagged with `chat_session_id` for traceability
-4. **Rich Tool Suite** — Five agent tools for explicit memory operations when needed
+4. **Rich Tool Suite** — Six agent tools for memory and document operations when needed
 
 ## Conclusion
 

package/docs/zh/openclaw.mdx

@@ -0,0 +1,96 @@
+---
+title: OpenClaw（中文）
+---
+
+通过 `memorylake-openclaw` 插件为 [OpenClaw](https://github.com/openclaw/openclaw) 代理添加长期记忆。你的代理在不同会话之间会忘记一切——这个插件会自动“观察”对话、提取重要信息，并在相关时把它们带回上下文，从而解决遗忘问题。
+
+## 概览
+
+{/*<Frame>
+  <img src="/images/openclaw-architecture.png" alt="OpenClaw MemoryLake Architecture" />
+</Frame>*/}
+
+该插件提供：
+1. **自动召回（Auto-Recall）** — 在代理回复前，将与当前消息匹配的记忆注入到上下文中
+2. **自动捕获（Auto-Capture）** — 在代理回复后，将本轮对话发送到 MemoryLake，由其判断哪些内容值得长期保存
+3. **代理工具（Agent Tools）** — 提供 5 个工具，便于在对话中显式进行记忆操作
+
+自动召回与自动捕获默认静默运行，无需手动配置即可生效。
+
+## 安装
+
+```bash
+openclaw plugins install memorylake-openclaw
+```
+
+## 设置与配置
+
+<Note>请从 [app.memorylake.ai](https://app.memorylake.ai) 获取 API key 和 project ID。</Note>
+
+在你的 `openclaw.json` 中添加：
+
+```json5
+// plugins.entries
+"memorylake-openclaw": {
+  "enabled": true,
+  "config": {
+    "apiKey": "${MEMORYLAKE_API_KEY}",
+    "projectId": "proj-..."
+  }
+}
+```
+
+## 代理工具
+
+代理在对话中可调用以下 5 个工具：
+
+| 工具 | 说明 |
+|------|------|
+| `memory_search` | 用自然语言搜索记忆 |
+| `memory_list` | 列出某个用户已存储的全部记忆 |
+| `memory_store` | 显式保存一条事实 |
+| `memory_get` | 通过 ID 读取一条记忆 |
+| `memory_forget` | 通过 ID 删除一条记忆 |
+
+## CLI 命令
+
+```bash
+# 搜索记忆
+openclaw memorylake search "what languages does the user know"
+
+# 查看统计信息
+openclaw memorylake stats
+```
+
+## 配置项
+
+| Key | 类型 | 默认值 | 说明 |
+|-----|------|--------|------|
+| `apiKey` | `string` | — | **必填。** MemoryLake API key（支持 `${MEMORYLAKE_API_KEY}`） |
+| `projectId` | `string` | — | **必填。** MemoryLake project ID |
+| `host` | `string` | `https://app.memorylake.ai` | MemoryLake 服务端点 URL |
+| `autoRecall` | `boolean` | `true` | 每轮对话前注入记忆 |
+| `autoCapture` | `boolean` | `true` | 每轮对话后存储事实 |
+| `topK` | `number` | `5` | 每次召回最多注入的记忆条数 |
+| `searchThreshold` | `number` | `0.3` | 最小相似度阈值（0–1） |
+| `rerank` | `boolean` | `true` | 对搜索结果重排以提升相关性 |
+
+## 关键特性
+
+1. **零配置** — 自动召回与自动捕获开箱即用，无需额外提示词或手动开关
+2. **异步处理** — 记忆提取通过 MemoryLake API 异步执行
+3. **会话追踪** — 对话会附带 `chat_session_id`，便于追溯与排查
+4. **工具完备** — 需要时可使用 5 个代理工具显式管理记忆
+
+## 总结
+
+`memorylake-openclaw` 插件让 OpenClaw 代理以极低成本获得持久记忆能力。你的代理可以跨会话自动记住用户偏好、事实与上下文，从而更稳定地连续协作。
+
+{/*<CardGroup cols={2}>
+  <Card title="MemoryLake" icon="brain" href="https://app.memorylake.ai">
+    MemoryLake platform
+  </Card>
+  <Card title="OpenClaw" icon="robot" href="https://github.com/openclaw/openclaw">
+    OpenClaw agent framework
+  </Card>
+</CardGroup>*/}