c0wrk

[!WARNING]
Early Alpha Stage — This project is under active development and not yet stable.
Features, APIs, and configuration formats may change without notice.
Use at your own risk.

A desktop AI agent for work that won't fit in a single reply — research, coding, writing, anything multi-step.

What makes c0wrk different

• Tackles genuinely hard, multi-part tasks. Rather than answering in one shot, c0wrk plans the work, takes it step by step, and replans when something goes wrong, so a complicated request gets carried all the way to a finished result.
• Goals that run to the finish. Start a message with /goal and c0wrk commits to a concrete, checkable definition of "done" that you sign off on up front — then keeps working turn after turn, checking its own progress, until it can prove the goal is met, it's stuck, the budget runs out, or you hit pause. No stopping at the first plausible-looking answer.
• A real desktop workspace. Chat sits next to a file tree, a code viewer with inline diffs, and live panels that show what the agent is doing and why.
• Two modes for two kinds of work. CODE points the agent at a real project: full toolset, in-place file edits, inline diffs, and git. CHAT is for when you want to reason, research, or draft without touching a codebase — each CHAT session gets its own isolated scratch workspace, code-modifying tools are switched off, and no git is required.
• You stay in control. Every potentially risky action can require your approval, with an optional AI "judge" that reasons about borderline calls so you're not drowned in prompts — or drowned in risk.
• Defense in depth. The agent stays inside your project workspace by default; any step outside it — reading or writing files elsewhere on disk — needs your explicit OK. Shell commands go through a hardened denylist (tighter still in CHAT mode), web fetches are SSRF-protected, and every safety check runs before an operation can auto-approve.
• Bring your own model. Works with Anthropic, OpenAI, ChatGPT, Gemini, LM Studio, and any OpenAI- or Anthropic-compatible endpoint. Set it up once, switch models per task.
• Connects to your tools. Plug in external services through the Model Context Protocol (MCP), extend behaviour with composable agent skills, and run shell commands directly in a built-in terminal.
• Knows your codebase. Semantic + keyword search means c0wrk finds the right file or symbol even when you describe it loosely, not just when keywords happen to match.
• Picks up where you left off. Full session history is persisted, so you can close the app and resume a task exactly as it was.
• Searches the live web. Pulls up-to-date information via Tavily when the task needs current knowledge.
• Plays nice with corporate networks. HTTP/HTTPS proxy support with custom CA certs and bypass lists.

Download & install

Prebuilt desktop builds are published on the GitHub Releases page. Builds are currently unsigned — see the per-OS notes below for the one-time workaround.

Each release bundles three artifacts:

The ONNX Runtime library and the embedding models ship inside every archive, so vector search works without an extra download on your end.

macOS (Apple Silicon)

1. Download c0wrk-desktop-macos-arm64.zip and unzip it.

2. The app is unsigned, so macOS Gatekeeper will block first launch. Clear the quarantine attribute:

   xattr -cr /path/to/c0wrk-desktop.app
   

   Alternatively, right-click c0wrk-desktop.app → Open on first launch, then confirm in the Gatekeeper dialog.

3. Launch c0wrk-desktop.app.

Linux (amd64)

1. Download c0wrk-desktop-linux-amd64.tar.gz and extract it:

   tar -xzf c0wrk-desktop-linux-amd64.tar.gz
   

2. Run the binary:

   ./c0wrk-desktop
   

3. If the binary can't find libonnxruntime.so, run it from the extraction directory or add that directory to LD_LIBRARY_PATH.

Runtime dependencies (end users only need these shared libraries, not the -dev packages):

# Ubuntu/Debian
sudo apt install libgtk-3-0 libwebkit2gtk-4.1-0

Windows (amd64)

1. Download c0wrk-desktop-windows-amd64.zip and extract it.
2. The app is unsigned, so Windows SmartScreen may warn "Windows protected your PC". Click More info → Run anyway.
3. Run c0wrk-desktop.exe.

On first launch c0wrk creates its config automatically, then — once all runtime dependencies are in place — opens a dialog to set up an LLM provider.

Contributing

Interested in hacking on c0wrk? Architecture, build instructions, and the release process live in CONTRIBUTING.md.

License

Licensed under the MIT License.

About

Built by the c0wrk team with warmth, love, and c0wrk.