An open-source coding agent for your terminal: run sessions in your repo or an isolated sandbox, on any model, extended with plugins.
Under active development; APIs may change.
pnpm add -g @codeworksh/cli
export OPENAI_API_KEY=...
codework run --provider openai --model gpt-5.6-luna "Inspect and fix the failing tests"Any other provider works the same way with its own key; list providers and models with codework models:
export ANTHROPIC_API_KEY=...
codework run --provider anthropic --model claude-sonnet-5-5 "Inspect and fix the failing tests"- Any model — OpenAI, Anthropic, Google, xAI, OpenRouter, and more via one catalog (
codework models). - Subscription sign-in —
codework auth login --openai-codexor--github-copilot. - Persistent sessions — resume any session by ID; state lives in a local database.
- Sandboxes — run locally or in
daytona,vercel,memory, orsqldbsandboxes. - Server mode —
codework serveowns sessions and sandboxes; clients attach with--server. - Plugins —
codework plugin add <pkg>adds tools, prompts, and sandbox drivers, per project or globally (-g).
See the CLI docs for every command and flag.
| Package | Description |
|---|---|
@codeworksh/cli |
The codework coding agent CLI. |
@codeworksh/harness |
Effect-powered agent runtime with persistent sessions and sandboxes. |
@codeworksh/plugin |
SDK for tools, prompt plugins, and sandbox drivers. |
@codeworksh/aikit |
Unified multi-provider LLM API with streaming, tools, and usage. |
Requires Node.js 24.14.1+ and pnpm.
git clone https://github.com/codeworksh/codework.git
cd codework
pnpm install
pnpm cli -- --help # run the CLI from source
pnpm exec vp check
pnpm exec vp run typecheckRead AGENTS.md for repository conventions, and open an issue for bugs or ideas.
The curiosity to build this project was inspired by Mario Zechner’s work on Pi-Mono.