mind — brain-like persistent memory for Codex via AGENTS.md (one Python file, zero deps, offline)

[Da7-Tech]

I built a persistent, self-organizing project memory that works with Codex after one explicit mind init. The command exports a guard-marked working-memory block into AGENTS.md while preserving existing content.

Introducing mind 🧠

Agents that follow the exported contract record selected durable project facts, recall related context from rough hints, and carry that context across sessions.

Between sessions, the deterministic Dreamer decays stale salience, prunes weak connections, promotes recurring clusters, and flags possible contradictions. Unused facts can move to the archive by design; this is not a rollback system.

Positioning vs the neighboring tools (Mem0, Letta, Graphiti, HippoRAG 2): see the honest scope comparison in the README.

One stdlib-only Python file, zero dependencies, zero API keys, fully offline, MIT. The same project graph can be exported to GEMINI.md and CLAUDE.md for agents that read those files.

https://github.com/Da7-Tech/mind

Feedback and adversarial testing are welcome.