I built Engram because every time I started a new Claude Code session, it forgot everything. Same questions, same mistakes, zero context. AI agents have Alzheimer.

Engram is a memory layer for AI agents. Store facts, preferences, and decisions. Search them with full-text search. Recall the most important ones for context injection. 5 lines of Python, zero config.

  from engram import Memory
  mem = Memory()
  mem.store("User prefers dark mode", importance=8)
  results = mem.search("dark mode")
  context = mem.recall(limit=10)

Local-first: SQLite, runs on your machine. No cloud, no API keys, no telemetry.

Zero config: pip install engram-core and go. No Docker, no Postgres, no vector DB.

MCP native: First-class Model Context Protocol support — plug into Claude Code, Cursor, or any MCP client.

Privacy: Your data never leaves your machine. MIT licensed.

I use it daily with Claude Code via an auto-recall hook — every new session starts with my important memories pre-loaded. No more "where were we?"

Built with: Python, SQLite FTS5, FastAPI, MCP SDK.

Website: https://engram-ai.dev

GitHub: https://github.com/engram-memory/engram

PyPI: pip install engram-core

Would love feedback. What memory features would you want for your agents?