openclaw-mem 1.0.3 → 1.2.1

package/HOOK.md

@@ -0,0 +1,125 @@
+---
+name: openclaw-mem
+description: "Persistent memory system - saves session context and injects history into new sessions"
+homepage: https://github.com/openclaw-mem
+metadata:
+  {
+    "openclaw":
+      {
+        "emoji": "🧠",
+        "events": ["command:new", "gateway:startup", "agent:bootstrap", "agent:response", "agent:stop", "tool:post", "user:prompt", "message"],
+        "requires": { "config": ["workspace.dir"] },
+        "install": [{ "id": "local", "kind": "local", "label": "Local installation" }],
+      },
+  }
+---
+
+# OpenClaw-Mem: Persistent Memory System
+
+A claude-mem inspired memory system for OpenClaw that automatically captures tool usage, generates summaries, and injects relevant context into new sessions.
+
+## Features
+
+- 🧠 **Persistent Memory** - Context survives across sessions
+- 📊 **Progressive Disclosure** - Shows index first, fetch details on demand
+- 🔍 **Hybrid Search** - Full-text + semantic search
+- 🤖 **AI Compression** - Automatic summarization of observations
+- ⚡ **Token Efficient** - Only loads what's relevant
+- 📡 **Real-time Capture** - Records messages as they happen
+
+## Events Captured
+
+- `gateway:startup` - Initialize memory system
+- `agent:bootstrap` - Inject historical context
+- `agent:response` - Capture assistant responses in real-time
+- `agent:stop` - Save session summary
+- `command:new` - Save session content before reset
+- `tool:post` - Capture tool usage
+- `user:prompt` - Capture user messages
+- `message` - Capture all messages
+
+## Configuration
+
+```json
+{
+  "hooks": {
+    "internal": {
+      "entries": {
+        "openclaw-mem": {
+          "enabled": true,
+          "observationLimit": 50,
+          "fullDetailCount": 5,
+          "compressWithLLM": true
+        }
+      }
+    }
+  }
+}
+```
+
+## Storage
+
+Data is stored in SQLite at `~/.openclaw-mem/memory.db`:
+- `sessions` - Session records
+- `observations` - Tool call observations
+- `summaries` - Session summaries
+
+## Real-time Monitoring
+
+```bash
+node ~/.openclaw/hooks/openclaw-mem/monitor.js
+```
+
+## Usage
+
+The memory system works automatically. To search manually:
+
+```bash
+# Search memory
+openclaw memory search "authentication"
+
+# View status
+openclaw memory status
+```
+
+## MCP Server (Model Context Protocol)
+
+OpenClaw-Mem 提供 MCP Server，让 AI 可以按需查询记忆：
+
+### MCP 工具
+
+| Tool | Description |
+|------|-------------|
+| `__IMPORTANT` | 显示 3 层工作流说明 |
+| `search` | 搜索记忆索引 |
+| `timeline` | 获取某条记录的上下文 |
+| `get_observations` | 获取完整详情 |
+
+### 启动 MCP Server
+
+```bash
+# stdio 模式（标准 MCP）
+node ~/.openclaw/hooks/openclaw-mem/mcp-server.js
+
+# HTTP API 模式（兼容模式）
+node ~/.openclaw/hooks/openclaw-mem/mcp-http-api.js
+```
+
+### HTTP API 端点
+
+```bash
+# 搜索
+curl "http://127.0.0.1:18790/search?query=database&limit=10"
+
+# Timeline
+curl "http://127.0.0.1:18790/timeline?anchor=123"
+
+# 获取详情
+curl -X POST "http://127.0.0.1:18790/get_observations" -d '{"ids":[123,124]}'
+```
+
+## Disabling
+
+```bash
+openclaw hooks disable openclaw-mem
+```

package/LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2024 OpenClaw Contributors
+Copyright (c) 2026 Aaron
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

package/MCP.json

@@ -0,0 +1,11 @@
+{
+  "mcpServers": {
+    "openclaw-mem-search": {
+      "type": "stdio",
+      "command": "node",
+      "args": ["mcp-server.js"],
+      "cwd": "${HOOK_ROOT}",
+      "description": "Memory search tools for OpenClaw-Mem"
+    }
+  }
+}

package/README.md

@@ -1,184 +1,195 @@
-# OpenClaw-Mem 🧠
+# OpenClaw-Mem
 
-> Give your OpenClaw AI agent persistent long-term memory
+A persistent memory system for [OpenClaw](https://github.com/openclaw) that automatically captures conversations, generates summaries, and injects relevant context into new sessions.
 
-[![npm version](https://badge.fury.io/js/openclaw-mem.svg)](https://www.npmjs.com/package/openclaw-mem)
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+## Features
 
-English | [中文](README_CN.md)
+- **Persistent Memory** - Context survives across sessions
+- **Progressive Disclosure** - Shows index first, fetch details on demand
+- **Hybrid Search** - Full-text + LIKE search with CJK support
+- **AI Compression** - Automatic summarization of observations
+- **Token Efficient** - Only loads what's relevant
+- **Real-time Capture** - Records messages as they happen
+- **MCP Compatible** - Model Context Protocol server included
+- **HTTP API** - REST API for memory queries
 
-OpenClaw-Mem automatically captures your conversations and makes them searchable, allowing your AI assistant to remember what you've discussed across sessions.
+## Installation
 
-## ✨ Features
+### As OpenClaw Hook
 
-- **🔄 Automatic Memory Capture** - Conversations are saved automatically
-- **🔍 Full-Text Search** - Search through your entire conversation history
-- **📊 Progressive Disclosure** - Efficient token usage with layered context
-- **🎯 Topic Detection** - Automatically indexes discussions by topic
-- **💾 Local Storage** - All data stays on your machine (SQLite)
-- **⚡ Zero Config** - Works out of the box
+```bash
+# Clone to OpenClaw hooks directory
+git clone https://github.com/wenyupapa-sys/openclaw-mem.git ~/.openclaw/hooks/openclaw-mem
+cd ~/.openclaw/hooks/openclaw-mem
+npm install
+```
 
-## 🚀 Quick Start
+### As npm Package
 
 ```bash
-# Install and setup (one command!)
-npx openclaw-mem init
-
-# Restart OpenClaw gateway
-openclaw gateway restart
+npm install openclaw-mem
 ```
 
-That's it! Start chatting and your conversations will be remembered.
+## Quick Start
 
-## 📖 Usage
+1. **Install the hook** (see above)
 
-### Recalling Memories in Chat
+2. **Run setup** - configure your DeepSeek API key (prompted automatically after install)
+   ```bash
+   # Or run manually later
+   npm run setup
+   # or
+   npx openclaw-mem-setup
+   ```
 
-Ask your AI assistant:
-- "What did we discuss before?"
-- "What were we working on last time?"
-- "Remind me about the authentication issue"
+3. **Restart OpenClaw** to load the hook
 
-### CLI Commands
+4. **Start chatting** - conversations are automatically saved
 
-```bash
-# Check memory status
-npx openclaw-mem status
+5. **Query memories** - ask "what did we discuss before?" and the AI will search the memory database
 
-# Search your memory
-npx openclaw-mem search "authentication"
-npx openclaw-mem search "AI memory"
+## Events Captured
 
-# Remove (keeps database)
-npx openclaw-mem uninstall
-```
+| Event | Description |
+|-------|-------------|
+| `gateway:startup` | Initialize memory system |
+| `agent:bootstrap` | Inject historical context |
+| `agent:response` | Capture assistant responses |
+| `agent:stop` | Save session summary |
+| `command:new` | Save session before reset |
+| `tool:post` | Capture tool usage |
+| `user:prompt` | Capture user messages |
 
-## 🏗️ How It Works
+## API Reference
 
-```
-+-------------+     +------------------+     +------------+
-|  Telegram   | --> | OpenClaw Gateway | --> |  AI Agent  |
-+-------------+     +------------------+     +------------+
-                            |                      |
-                            v                      v
-                    +--------------+        +------------+
-                    | openclaw-mem |        | Read Tool  |
-                    |    hook      |        +------------+
-                    +--------------+               |
-                            |                      v
-        +-------------------+----------------------+
-        |                   |
-        v                   v
-+---------------+   +------------------+
-|   SQLite DB   |   | SESSION-MEMORY.md|
-|  (persistent) |   |    (injected)    |
-+---------------+   +------------------+
+### HTTP API (Port 18790)
+
+```bash
+# Search memories
+curl -s -X POST "http://127.0.0.1:18790/search" \
+  -H "Content-Type: application/json" \
+  -d '{"query":"keyword","limit":10}'
+
+# Get observation details
+curl -s -X POST "http://127.0.0.1:18790/get_observations" \
+  -H "Content-Type: application/json" \
+  -d '{"ids":[123,124]}'
+
+# Get timeline context
+curl -s -X POST "http://127.0.0.1:18790/timeline" \
+  -H "Content-Type: application/json" \
+  -d '{"anchor":123}'
+
+# Health check
+curl "http://127.0.0.1:18790/health"
 ```
 
-### Event Flow
+### Shell Scripts
 
-1. **`gateway:startup`** - Initializes the memory database
-2. **`agent:bootstrap`** - Injects historical context into new sessions
-3. **`command:new`** - Saves session summary when you start fresh
+```bash
+# Search (handles CJK encoding automatically)
+~/.openclaw/hooks/openclaw-mem/mem-search.sh "关键词" 10
 
-### Progressive Disclosure
+# Get details
+~/.openclaw/hooks/openclaw-mem/mem-get.sh 123 124 125
+```
 
-To optimize token usage, memories are organized in layers:
+### MCP Server
 
-| Layer | Content | Tokens |
-|-------|---------|--------|
-| Index | Compact table of all observations | ~Low |
-| Topics | Key discussion summaries | ~Medium |
-| Details | Full content (on demand) | ~High |
+```bash
+# Start MCP server (stdio mode)
+node mcp-server.js
+```
 
-## 📁 File Locations
+MCP Tools:
+- `search` - Search memory index
+- `timeline` - Get context around an observation
+- `get_observations` - Fetch full details
 
-| File | Location | Purpose |
-|------|----------|---------|
-| Database | `~/.openclaw-mem/memory.db` | Persistent storage |
-| Hook | `~/.openclaw/hooks/openclaw-mem/` | Event handlers |
-| Memory Context | `~/.openclaw/workspace/SESSION-MEMORY.md` | Injected context |
+## Configuration
 
-## ⚙️ Configuration
+### Environment Variables
 
-### Customize MEMORY.md
+```bash
+# Required for AI summarization (optional but recommended)
+export DEEPSEEK_API_KEY="your-deepseek-api-key"
 
-Add your preferences to `~/.openclaw/workspace/MEMORY.md`:
+# Optional: Custom DeepSeek endpoint
+export DEEPSEEK_BASE_URL="https://api.deepseek.com/v1"
 
-```markdown
-## User Preferences
-- Communication style: Technical and concise
-- Language: English
+# Optional: Custom model
+export DEEPSEEK_MODEL="deepseek-chat"
+```
 
-## Long-Term Context
-- Working on: AI memory systems
-- Interests: Machine learning, distributed systems
+Get your DeepSeek API key at: https://platform.deepseek.com/
 
-## Ongoing Focus
-- Current project: Building a chat application
-```
+> **Note:** Without `DEEPSEEK_API_KEY`, the system will still work but won't generate AI summaries for sessions.
 
-### Dynamic Topic Detection
+### OpenClaw Config
 
-The system automatically extracts topics from your actual conversations:
-- Concepts mentioned frequently are detected automatically
-- No hardcoded keywords - everything comes from your data
-- Topics evolve as your discussions change
+Add to your OpenClaw config:
 
-## 🔧 Troubleshooting
+```json
+{
+  "hooks": {
+    "internal": {
+      "entries": {
+        "openclaw-mem": {
+          "enabled": true,
+          "observationLimit": 50,
+          "fullDetailCount": 5
+        }
+      }
+    }
+  }
+}
+```
 
-### Memory not being recalled?
+## Storage
 
-1. **Restart the gateway:**
-   ```bash
-   openclaw gateway restart
-   ```
+Data is stored in SQLite at `~/.openclaw-mem/memory.db`:
 
-2. **Check hook is installed:**
-   ```bash
-   ls ~/.openclaw/hooks/openclaw-mem/
-   ```
+| Table | Description |
+|-------|-------------|
+| `sessions` | Session records |
+| `observations` | Tool calls and messages |
+| `summaries` | Session summaries |
+| `user_prompts` | User inputs |
 
-3. **Verify database has content:**
-   ```bash
-   npx openclaw-mem status
-   ```
+## Development
 
-### Search not finding results?
+```bash
+# Run tests
+npm test
 
-- Try simpler search terms
-- Check for typos
-- Use keywords from the actual conversation
+# Start HTTP API server
+npm run api
 
-### Database issues?
+# Start MCP server
+npm run mcp
 
-```bash
-# Reset database (⚠️ deletes all memories)
-rm ~/.openclaw-mem/memory.db
-openclaw gateway restart
+# Monitor real-time activity
+node debug-logger.js
 ```
 
-## 🤝 Contributing
+## 3-Layer Retrieval Workflow
+
+For efficient token usage, use progressive disclosure:
 
-Contributions are welcome! Please feel free to submit a Pull Request.
+1. **Search** → Get index with IDs (~50-100 tokens/result)
+2. **Timeline** → Get context around interesting results
+3. **Get Observations** → Fetch full details ONLY for filtered IDs
 
-1. Fork the repository
-2. Create your feature branch (`git checkout -b feature/amazing-feature`)
-3. Commit your changes (`git commit -m 'Add amazing feature'`)
-4. Push to the branch (`git push origin feature/amazing-feature`)
-5. Open a Pull Request
+This approach saves ~30% tokens compared to fetching everything.
 
-## 📄 License
+## License
 
-MIT License - see [LICENSE](LICENSE) for details.
+MIT
 
-## 🙏 Acknowledgments
+## Contributing
 
-- Inspired by [claude-mem](https://github.com/anthropics/claude-mem)
-- Built for [OpenClaw](https://openclaw.ai)
+Pull requests welcome! Please ensure tests pass before submitting.
 
----
+## Credits
 
-<p align="center">
-  Made with ❤️ for the AI community
-</p>
+Inspired by [claude-mem](https://github.com/anthropics/claude-code) plugin architecture.