Skip to content
HearthworkHearthwork

Your own GPU, your coding agents

Hearthwork: local models for your coding agents, tuned to your own machine

Run a local AI model on your own computer and connect your coding agent to it, from one menu. Private, no per-token cost, tuned to your hardware.

Get startedGitHub

Why Hearthwork

Private

The model runs on your own GPU and RAM. Your code and prompts stay on your computer.

No per-token cost

Once the model is downloaded, using it costs electricity, not an API bill.

Works with the agents you already use

Claude Code, Codex, OpenCode, Qwen Code and Aider, each started with one command.

Supported agents

Hearthwork connects each one to your local model without touching its own settings.

Claude CodeCodexOpenCodeQwen CodeAider

How it works

One host per computer owns the model and a single queue. Everything goes through a relay that cleans up requests and repairs tool calls.

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

Read how it works

Use it anywhere

Run the model on the computer with the GPU and use it from your laptop, at home or away.

Share on your network

Pair another computer with a 6-digit PIN. It needs no model and no setup. Learn more

Tailscale

Reach your model from anywhere over an encrypted private network. Learn more

Mobile app

In development, Android first. Learn more

Everything you need around the model

Model check from your PC

Reads your hardware and tells you honestly whether a local model is worth it, and which ones fit.

Doctor

Tests the server, the API and your agent step by step and says which layer is at fault.

Diagnose

Explains what slowed or broke your sessions, using numbers measured on your own requests.

Budgets and presets

Leave room for gaming or work: cap VRAM, slots and threads, or unload when idle.

Setup profiles

Save a setup that works, with proof, then re-check, export or restore it.

Automatic recovery

If the model server crashes, the host restarts it and tells clients clearly if it cannot.

API mode

One local endpoint that speaks both the OpenAI and the Anthropic API, for any other tool.

Subagents and MCP

Let another agent hand small coding tasks to your local model.

Quick start

Terminal window
uvx hearthwork check          # is this computer suited, and which models fit? (installs nothing)
uv tool install hearthwork    # install (or: pipx install hearthwork)
hearthwork model unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF   # download a model that fits
cd my-project
hearthwork claude             # run an agent on the local model (also: codex, opencode, qwen, aider)

The first run sets things up for you: hardware check, llama.cpp and a models folder. Start here

See it run

Hearthwork demo: pick Claude Code, pick a model, it loads, the agent reads a file and answers; then Codex on the same model; then quit

A replay of a real run. Waits are sped up, the agent screens are simplified, and paths are shortened.