How it works

How Hearthwork works: menu, first-run setup, model folder, llama-server using GPU and RAM, and a relay between the agents and the model

One host per computer

hearthwork start, or any command that needs the model, starts the Hearthwork host in the background. The host owns llama-server and the single relay and request queue that every client goes through: agents, task and MCP, hearthwork api and hearthwork share.

It listens on 127.0.0.1:8470. Each agent session identifies itself with a token like local-claude-ab12.
The model keeps running between agent sessions, and nothing opens a window.
A model started by an older Hearthwork is adopted, not restarted.

The pieces

The relay: the queue, request cleanup and tool-call repair.
Context and models: how context is sized, how layers are placed on the GPU, and how models are recommended.
Faster first messages. Each agent’s processed system prompt is saved to cache/slots/ when a session ends or the server stops, and restored when the server starts.
Settings live in the server section of config.json. hearthwork setup recalculates them, and hearthwork start --model X or --context N override a single run.
Recovery keeps the model server running. See Recovery.