Memory LanceDB Context Engine

A smart ContextEngine plugin for OpenClaw that integrates with memory-lancedb-pro for retrieval-augmented context management, auto-capture, intelligent memory injection, batch import, and memory-aware compaction.

Features

🧠 Intelligent Memory Injection

• Automatically retrieves relevant memories during context assembly
• Injects historical context into system prompt with relevance scores
• Supports hybrid search (vector + BM25 + reranking)

📝 Auto-Capture

• Detects important user messages using pattern matching
• Supports Chinese, English, and Czech trigger phrases
• Categories: preference, decision, fact, entity, other

📦 Batch Import (ingestBatch)

• Import multiple messages at once
• Configurable batch size (default: 100)
• Automatic duplicate detection
• Progress tracking with detailed results

🔄 Historical Session Bootstrap

• Import historical sessions into memory on startup
• Archive file support (.archive, .bak, history-*)
• Preserves context across sessions

🗜️ Custom Compaction Strategies

| Strategy | Description | |----------|-------------| | aggressive | Maximum compression, minimal preservation | | balanced | Default, good balance between compression and context | | conservative | Minimal compression, maximum preservation | | custom | User-defined instructions |

⚡ Smart Summarization

• Auto-generates summaries from recent messages
• Extracts decisions, facts, entities, preferences
• Stores summaries for future context recovery

Installation

# Install via npm
npm install memory-lancedb-context

# Or clone to your OpenClaw plugins directory
git clone https://github.com/2951461586/memory-lancedb-context.git
cd memory-lancedb-context
npm install

Configuration

Add to your openclaw.json:

{
  "plugins": {
    "slots": {
      "memory": "memory-lancedb-pro",
      "contextEngine": "memory-lancedb-context"
    },
    "entries": {
      "memory-lancedb-pro": {
        "enabled": true,
        "config": {
          "embedding": {
            "apiKey": "your-api-key",
            "model": "text-embedding-3-small"
          }
        }
      },
      "memory-lancedb-context": {
        "enabled": true,
        "config": {
          "autoCapture": true,
          "enableMemoryInjection": true,
          "maxMemoriesToInject": 5,
          "minInjectionScore": 0.4,
          "compactionStrategy": "balanced",
          "enableDuplicateDetection": true,
          "duplicateThreshold": 0.92,
          "enableSmartSummarization": true,
          "maxBatchSize": 100
        }
      }
    }
  }
}

Configuration Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | autoCapture | boolean | true | Auto-capture user messages that match memory triggers | | enableMemoryInjection | boolean | true | Inject relevant memories into context during assemble | | maxMemoriesToInject | integer | 5 | Maximum number of memories to inject (1-10) | | minInjectionScore | number | 0.4 | Minimum relevance score for memory injection (0-1) | | defaultAgentId | string | "main" | Default agent ID for scope resolution | | enableMemoryCompaction | boolean | true | Enable memory-based intelligent compaction | | preserveImportantTurns | boolean | true | Store important turns to memory during compaction | | maxPreservedTurns | integer | 20 | Maximum recent turns to analyze (5-50) | | compactionStrategy | string | "balanced" | Compaction strategy (aggressive/balanced/conservative/custom) | | customCompactionInstructions | string | "" | Custom instructions for custom strategy | | enableMemoryDecay | boolean | false | Enable memory decay (future feature) | | memoryDecayDays | integer | 30 | Memory decay threshold in days | | enableDuplicateDetection | boolean | true | Enable duplicate detection | | duplicateThreshold | number | 0.92 | Similarity threshold for duplicates (0.8-1) | | enableSmartSummarization | boolean | true | Enable smart summarization | | maxBatchSize | integer | 100 | Maximum batch size for ingestBatch (10-500) |

API

ingestBatch(params)

Batch import messages into memory.

const result = await contextEngine.ingestBatch({
  messages: [
    { role: "user", content: "I prefer American coffee without sugar" },
    { role: "user", content: "Project codename is Phoenix" },
  ],
  scope: "agent:main",  // optional
  skipDuplicates: true,  // default: true
});

// Result:
{
  total: 2,
  ingested: 2,
  skipped: 0,
  failed: 0,
  details: [
    { text: "I prefer American coffee...", status: "ingested" },
    { text: "Project codename is Phoenix", status: "ingested" },
  ]
}

bootstrap(params)

Import historical session on startup.

const result = await contextEngine.bootstrap({
  sessionId: "session-123",
  sessionFile: "/path/to/session.json",
});

// Result:
{
  bootstrapped: true,
  reason: "Imported 15 messages from session, 5 from archives",
  data: {
    totalMessages: 50,
    ingested: 15,
    skipped: 35,
    failed: 0,
    importedArchives: 5
  }
}

Compaction Strategies

// Aggressive - maximum compression
{
  "compactionStrategy": "aggressive"
}

// Balanced - default
{
  "compactionStrategy": "balanced"
}

// Conservative - minimal compression
{
  "compactionStrategy": "conservative"
}

// Custom - user-defined
{
  "compactionStrategy": "custom",
  "customCompactionInstructions": "Focus on technical decisions. Preserve all code snippets."
}

How It Works

┌─────────────────────────────────────────────────────────────┐
│                     User sends message                       │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  ingest() → Detect triggers → Auto-store important info     │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  assemble() → Retrieve memories → Inject into system prompt │
│  <relevant-memories>                                        │
│  [HISTORICAL CONTEXT]                                       │
│  - [preference:agent:violet] User likes Americano...        │
│  - [decision:agent:violet] Project codename Phoenix...      │
│  </relevant-memories>                                       │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│                    Model generates response                  │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  afterTurn() → Smart summary → Store for future context     │
└─────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────┐
│  compact() → Strategy-based compression → Memory preserved  │
└─────────────────────────────────────────────────────────────┘

Memory Triggers

The plugin automatically captures messages containing:

English

• "remember", "prefer", "I like/hate/want"
• "we decided", "switch to", "migrate to"
• "important", "always", "never"
• Phone numbers, email addresses

Chinese

• "记住", "记一下", "别忘了"
• "偏好", "喜欢", "讨厌"
• "决定", "改用", "以后用"
• "我的...是", "重要", "关键"

Czech

• "zapamatuj si", "preferuji", "radši"
• "rozhodli jsme", "budeme používat"

Comparison with Legacy ContextEngine

| Feature | Legacy | Memory-LanceDB-Context | |---------|--------|------------------------| | ingest | ❌ no-op | ✅ Auto-capture important messages | | ingestBatch | ❌ N/A | ✅ Batch import with duplicate detection | | assemble | ❌ pass-through | ✅ Retrieve & inject memories | | bootstrap | ❌ no-op | ✅ Historical session import | | compact | ✅ LLM summarization | ✅ Strategy-based + memory preservation | | afterTurn | ❌ no-op | ✅ Smart summarization + storage |

Dependencies

• OpenClaw >= 2026.3.7
• memory-lancedb-pro - Required for memory storage

Development

# Clone the repository
git clone https://github.com/2951461586/memory-lancedb-context.git
cd memory-lancedb-context

# Install dependencies
npm install

# Test
npm test

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'feat: Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Acknowledgments

• OpenClaw - The AI agent framework this plugin is designed for
• LanceDB - Serverless vector database
• memory-lancedb-pro - The memory storage backend