Peer-to-peer memory sync for AI agents
When one agent learns something, every other agent on every machine gets it, automatically, with no central server in the middle.
How the sync works
Each device keeps an append-only journal of everything your agents have learned. Devices compare what they have using a Merkle index and exchange only the entries the other is missing. That keeps sync fast even as the history grows.
It runs over your private Tailscale network, so traffic is encrypted and never leaves machines you control. No extra infrastructure to stand up.
Always converging
Because the journal is a conflict-free set, the order entries arrive in does not matter. Every device converges on the same facts. You can add a machine, work offline, and rejoin without anything breaking.
Frequently asked
How does HiveMind sync without a server?
Devices talk directly to each other over Tailscale. Each compares a Merkle index of its journal with its peers and pulls only the missing entries. There is no central server or broker.
Is sync efficient over time?
Yes. Only the differences between devices are transferred, not the whole history, so sync stays fast as your memory grows.