Commands at a glance

For every flag and its exact help text, see the generated command reference.

Command	What it does
hearthwork	The menu.
hearthwork claude|codex|opencode|qwen|aider [args]	Run that agent on the local model. Add --dry-run first to show what would run.
hearthwork agents	Supported agents: installed, version, API, how each connects.
hearthwork task "TASK"	One coding task for another agent to delegate (no terminal needed).
hearthwork mcp [install claude|codex|print]	MCP server so any agent can use the local model as a subagent.
hearthwork start [--model X] [--context N] [--force] [--over-budget]	Start the model under the host, or switch it.
hearthwork stop [--force]	Save the prompt cache, then stop the model and the host.
hearthwork serve [--model X]	Run the host in this terminal with its log. Ctrl+C stops it.
hearthwork idle [30|off]	Unload the model after N idle minutes (off by default).
hearthwork budget [show|use dedicated|balanced|work|set ...|reset]	Resource budget.
hearthwork context [model] [N|auto]	Context range per model; save a size for a model.
hearthwork status [--json]	Host, model, context, slots, requests per client, budget, switch, idle and recovery state.
hearthwork logs [--server] [-f]	The host’s log. --server shows llama-server’s output.
hearthwork doctor [--agent X] [--quick] [--json]	Compatibility check. Says which layer is at fault.
hearthwork diagnose [--last N | --since 30m] [--client X] [--json] [--export FILE]	What limited your sessions, from measured numbers.
hearthwork check [--json]	Is this computer suited, and which models fit.
hearthwork model <link>	Download a GGUF model from Hugging Face.
hearthwork profile save|list|show|check|apply|export|import	Verified setup profiles.
hearthwork share [--port N] [--allow-public] [--tailscale]	Share the running model with other computers.
hearthwork connect [host[:port]]	Use the model of a computer that is sharing.
hearthwork disconnect	Back to local models.
hearthwork devices [remove <name>]	Trusted devices. remove revokes one.
hearthwork api [--port N] [--show]	OpenAI + Anthropic compatible endpoint on this computer.
hearthwork alias add|list|remove	Short commands for agents.
hearthwork bench [--agent X] [--all] [--warmup] [--show]	Graded coding tasks through an agent, with a scoreboard.
hearthwork tune [--time 30m] [--min-context N] [--dry-run] [--json]	Try a few installed configurations, save the best tested one as a profile.
hearthwork runtime [install <name>|use <name>|update|rollback]	llama.cpp engines (CUDA, ROCm, Vulkan, SYCL, Metal, CPU) side by side.
hearthwork setup [--update-llama] [--reset] [--import FOLDER]	Hardware check, llama.cpp download or update, models folder.
hearthwork update	Update Hearthwork itself.
hearthwork --version	Version and data folder.