Engram

What it does

Provides a reliable, local memory system for AI agents using an event-sourced architecture. It stores 'episodes' of interaction and allows the agent to retrieve them using vector (semantic), keyword, or hybrid search, ensuring that agent memory remains private and fast.

Tools

• update_episode: Update or soft-delete an existing memory episode.
• Semantic Search: Retrieve relevant memories based on meaning using local embeddings.
• Keyword Search: Find memories using exact word matches.
• Hybrid Search: Combine semantic and keyword search for high-precision retrieval.

Installation

Add the following to your claude_desktop_config.json:

{
  "mcpServers": {
    "engram-memory": {
      "command": "/absolute/path/to/engram",
      "args": ["stdio"],
      "env": {
        "ENGRAM_SERVER_URL": "http://localhost:3490"
      }
    }
  }
}

Note: Requires engram serve running in the background and Ollama with nomic-embed-text.

Supported hosts

• Claude Desktop
• Claude Code
• Cursor