It is deliberately small — write a message, read the recent ones back, and move between threads. What separates it from a local memory file is where the data goes: onto shared storage with a browsable hub, so the same history can follow an agent across machines.
A small server over the Membase memory layer. Messages you save go to the Unibase network, and the conversation they belong to preloads its history when the server starts.
- The current conversation identifier, so the agent knows which thread it is writing to (get_conversation_id)
- A switch to a different conversation without restarting (switch_conversation)
- A message or memory written into the current conversation (save_message)
- The last few messages read back out (get_messages)
- Stored messages viewable on the Membase hub in a browser
An account on the network to upload under, an instance identifier, and a conversation identifier unique to you — that last one decides which history preloads. All three are supplied as environment settings when the client launches the server. Python with uv.
Build from source — clone the repository and build it, then point your client at the binary