Welcome to MemMesh
MemMesh is a persistent memory engine for AI agents. It gives your agents durable memory that survives across sessions, tools, and machines — so an agent that learned something yesterday still knows it today, in any editor it runs in.
Where most “AI memory” stops at recall, MemMesh adds two things on top:
- Prediction — it mines observed behavior into patterns and forecasts the next likely event with a calibrated confidence score.
- Provenance & governance — every recalled fact can be traced back to when and how it was learned, with GDPR-style export and erasure built in.
MemMesh is open source and private by default — the engine can run entirely on your own infrastructure. See Self-Hosting.
How it works at a glance
- Observe — you feed the engine raw text (a conversation turn, a decision, a fact). It decides what’s worth keeping.
- Learn — observations are distilled into structured memories, entities, and patterns over time.
- Recall — agents search and retrieve relevant memories before they plan or act.
This loop is the core of MemMesh and is covered in Lifecycle.
Get set up in one command
The local engine wires into your editor over MCP with a single command:
npx @thinkfleet/memmesh installNo API key, no server, no Docker — memory is stored in a SQLite file on your machine. See the Quickstart for the full walkthrough.