An AI coding agent that works with the model you choose.

Handwork gives an AI model the tools to work on a real project: search the code, edit files, and run commands. Ask it to fix a bug, explain how something works, or make a change and run the tests.

Saved conversations, project instructions, skills, and MCP tools let you pick up work across sessions.

The agent runtime and terminal interface are written in Zig. The goal is a small memory footprint and a terminal that stays responsive during long sessions.

A JavaScript SDK lets other apps use the same agent with their own interface and tools.

The benchmark results include every attempt, including failures, with timing, token use, and memory. Models and run dates differ between agents, and four tasks do not establish an overall ranking.

Connect a cloud provider with an API key or subscription, or point Handwork at your own Ollama server. The Codex and Copilot adapters run those providers' own agent engines.

Handwork can edit files and run programs. Set permissions before your first task, then review the diff and test results.