Offline-first memory for AI agents
Agents keep working with no connection. When a device comes back, its entries merge cleanly, and no device is ever blocked waiting on another.
Local-first, always available
Every device holds the full memory and works against it directly. An agent does not pause when the network drops; it keeps reading and writing its local journal exactly as before. Availability is never at the mercy of a server or a peer being reachable.
That is the practical payoff of having no central store: there is nothing to be offline from. The system is the set of devices you run, and any one of them is complete on its own.
Clean convergence on reconnect
When a device rejoins, it exchanges only the entries the others are missing and they exchange the ones it is missing. Because the journal is a conflict-free set, the order things arrive in does not matter, so two devices that worked apart for a day merge into one consistent memory without manual reconciliation.
Frequently asked
Do my agents need a connection to work?
No. Each agent reads and writes its local memory whether or not it can reach any peer. When a device reconnects, its entries merge with the others. Nothing blocks waiting on the network.
What happens to changes made offline?
They are kept in the local append-only journal and exchanged the next time devices sync. Because the journal is a conflict-free set, the order entries arrive in does not matter and every device converges on the same facts.