Brain file format for AI agents – one file, any LLM, sub-millisecond queries

1 point

4 months ago

Every AI agent has amnesia. Claude doesn't remember your last conversation. GPT doesn't know what you decided last week. The current solutions — vector databases, markdown files, key-value stores — lose structure, can't track reasoning chains, and lock you to one provider.

I built AgenticMemory: a binary graph format where every cognitive event (facts, decisions, inferences, corrections) is a node with typed edges (caused_by, supports, supersedes). One .amem file holds your agent's entire knowledge graph. Works with any LLM.

Key numbers: • 276ns to add a node • 3.4ms to traverse 5 levels deep in a 100K-node graph • 9ms similarity search across 100K nodes • ~24 MB for a year of daily use • A lifetime of memory fits in under 1 GB

Built in Rust. Zero dependencies. Python SDK: pip install agentic-brain. Rust CLI: cargo install agentic-memory.

https://github.com/agentic-revolution/agentic-memory