Engram

What is Engram?

Engram is a memory layer for AI assistants—a local server that stores, searches, and manages contextual information for tools like Claude Desktop and Claude Code. It provides a hybrid recall algorithm, automatic secret detection, and multiple interfaces (CLI, REST API, MCP, and web dashboard) to give AI agents persistent, long-term memory.

How to use Engram?

Clone the repository, run npm install, then start the server with npm start. Engram can be used via CLI commands (engram remember, engram recall, engram list, engram status), integrated with Claude Desktop by adding an MCP entry in claude_desktop_config.json, accessed through a REST API at http://localhost:3838, or managed through the web dashboard at http://localhost:5173.

Key features of Engram

• Hybrid recall algorithm using similarity, recency, confidence, access, and FTS
• Five memory categories: preference, fact, pattern, decision, outcome
• Automatic secret detection and redaction for common credential types
• Memory consolidation: duplicate removal, contradiction detection, decay
• Namespace organization for projects or contexts
• PM2 support for persistent background service

Use cases of Engram

• Remember user preferences like coding style or framework choices
• Document project decisions and infrastructure facts for context-aware code generation
• Store and recall recurring workflow patterns (e.g., "always runs tests before commit")
• Maintain a persistent memory of outcomes across multiple development sessions

FAQ from Engram

How do I install Engram?

Clone the repository, run npm install, then start the server with npm start. The REST API runs on port 3838 and the dashboard on port 5173.

How do I integrate Engram with Claude Desktop?

Add an MCP server entry to ~/Library/Application Support/Claude/claude_desktop_config.json with the command node /path/to/engram/bin/engram.js start --mcp-only. Alternatively, use the Integration Wizard in the web dashboard at http://localhost:5173.

Does Engram require an internet connection?

Yes, on first use it downloads a 23MB embedding model to ~/.engram/models/. After that, all operations are local.

How do I resolve a database lock error?

Close all connections and restart the process: pkill -f "node bin/engram.js" then node bin/engram.js start. SQLite WAL mode normally prevents locks.

What is the performance of Engram?

Memory storage takes ~1ms per memory. Recall search takes ~10-50ms for 5 results. Database size is ~2KB per memory. Base memory usage is ~50MB plus the loaded model.