AI coding agents forget everything when the session ends, and switching agents means explaining it all again. Graph-MIND is a shared memory that lives on your own PC. Whichever agent or model you run, they all read and write the same memory. Decide something with one agent on Monday, ask another on Wednesday, and it knows.
- Saves conversations word for word. No summarizing or extraction, and no model call when saving (zero tokens). Saving is a single database write.
- Gives back only what's needed. When a question depends on the past, the agent asks over MCP and gets back the few thousand tokens that answer it. Korean and English.
- Works with any model. One install connects Claude Code, Codex, Cursor, Gemini CLI and more. Through OpenCode, any model that can call tools (DeepSeek, Qwen, GLM, …) shares the same memory.
- Yours. No account, no cloud, no telemetry. Passwords and keys are masked before anything is stored, and "delete the password I typed earlier" deletes it.
- Across your PCs. Paste the code the AI gives you into the AI on another PC, and both use the same memory. Connects over your office network or Tailscale, encrypted in transit.
Results: 88.8% on all 500 LongMemEval questions. On 40 questions fixed in advance, with the same answer model and the official judge: Graph-MIND 85.0%, MemPalace 57.5%, Mem0 (open source) 52.5%. In that comparison Mem0 spent about 640,000 tokens per question on saving. Every question, answer and verdict is published in the repo.
Install
pip install graph-mind-memory && graph-mind-install
On Windows, just double-click the installer.
Still alpha: used daily on Windows; macOS and Linux pass CI but haven't seen daily use yet. License: AGPL-3.0. Feedback welcome.