HiveMind proposes shared memory for Hermes Agent teams
The Reddit post says separate Hermes Agent setups can leave useful corrections and discoveries on one person's machine. HiveMind stores each member's memories in a JSONL Event Log, syncs them between machines with Syncthing, and rebuilds a local SQLite database. Hermes then reads that database through FTS5 search.
Key points
- HiveMind aims to let Hermes Agent users share corrections and discoveries across a team.
- Each member writes memories to a JSONL Event Log.
- Syncthing syncs those logs directly between machines.
- A Merge Engine replays the logs and rebuilds a local SQLite database.
- The proposed setup avoids a server, API, and cloud service.
Quick term guide
- Hermes Agent
- It appears to be a tool or community for building and managing AI agents.
- JSONL Event Log
- A file that stores one record per line as a running history.
- Syncthing
- A tool that keeps files matched across different machines.
- SQLite database
- A small file-based place for an app to store data.
- database
- A large collection of organized data used for search and analysis.
- hermes-agent
- A likely name for Nous Research’s agent-style AI tool or service.
- edge cases
- Unusual or unexpected inputs that fall outside the normal, expected use of a product.
- edge case
- An unusual or unexpected situation that falls outside the normal flow and often causes errors