We built ate because AI agents keep forgetting everything between sessions. RAG retrieves documents, but agents need to remember what they learned — decisions, preferences, context.

ate stores thousands of memories in a single .mv2 file (~2MB for 10K memories). It's encrypted (AES-256), works offline, and agents can carry their memory file anywhere.

Key features: - Multi-agent concurrent access (MVCC, no locks) - Memory Capsules for sharing context between agents - Trains of thought for reasoning chains - Works with CrewAI, LangChain, and Agno (PRs open)

Install: pip install ate-memory Try: ate memory init my-agent && ate memory add my-agent --text "first memory"

We're using it to give our own agents persistent memory. Would love feedback from anyone building multi-agent systems.