Hermes Fleet Memory: memory tool for Hermes Agent

Shared vector memory for Hermes agent fleets with on-demand MCP retrieval and work/personal firewalls

What Hermes Fleet Memory does

The project runs as a FastMCP server that exposes 12 tools for vector-backed memory retrieval and remote execution, with Qdrant listed as the vector database. Hermes is configured with memory.provider set to none, so nothing is added to the system prompt ambiently, and the agent calls tools such as fleet_memory_search only when a question needs it. Facts are stored in named slots using deterministic UUID5 identifiers and last-write-wins timestamps, so an updated setting overwrites the old one instead of creating a conflicting duplicate.

Privacy is handled by a client-level gate, FLEET_HARD_DOMAIN, which is set to work, personal or all. The README says this blocks work data from reaching a personal machine in code, rather than relying on the model to comply. A standalone mode needs no remote VPS, Qdrant cluster or reverse tunnels, and a --doctor flag checks the node. The same server can be registered in Claude Desktop, Cursor and Claude Code, and the project also describes a multi-node mesh with a central hub.

Key features

12 FastMCP tools for on-demand memory search and remote execution
No ambient prompt memory: Hermes runs with memory.provider set to none
In-place updates through UUID5 slot IDs and last-write-wins timestamps
FLEET_HARD_DOMAIN firewall with work, personal and all domains
Standalone mode with a non-interactive setup wizard
Doctor command that checks a node and always exits

When to use it

Sharing what a Hermes agent learned on a work laptop with the same agent on a home PC
Keeping work and personal memories apart on different machines
Giving a Telegram-connected Hermes bot the same memory as a desktop client

Who it is for: People who run Hermes Agent or other MCP clients on several machines and want one shared memory without extra prompt cost.

How it fits with Hermes Agent

Built for Hermes Agent, which connects to it as an MCP server in config.yaml. It also works with Claude Desktop, Cursor and Claude Code.

How to install Hermes Fleet Memory

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

git clone https://github.com/amrlazw/hermes-fleet-memory.git
cd hermes-fleet-memory
python setup.py --role standalone --non-interactive
python client/fleet_memory.py --doctor

Requirements: Python and an MCP client; standalone mode needs no remote VPS, Qdrant cluster or reverse tunnels

FAQ

What is Hermes Fleet Memory?

Hermes Fleet Memory is a FastMCP server that gives Hermes Agent and other AI clients one shared, searchable vector memory across devices, with work and personal domains kept separate.

Does Hermes Fleet Memory work with Hermes Agent?

Yes. Add it under mcp_servers in ~/.hermes/config.yaml with the path to client/fleet_memory.py. The README gives equivalent setups for Claude Desktop and Claude Code.

How do I install Hermes Fleet Memory?

Clone the repository, run python setup.py --role standalone --non-interactive, then check the node with python client/fleet_memory.py --doctor. Register the server in your AI client afterward.