Skip to main content
Coding agents like Claude Code, Codex CLI, Gemini CLI, Cursor, Devin, and Zed can use Autonomy documentation to ship complete autonomous applications.
  1. Follow the initial instructions to Get Started with Autonomy.
  2. Set up your favorite coding agent and give it the following prompt.
Prompt:
Use curl to fetch the raw markdown from this doc and all its linked pages on creating a new Autonomy application:
https://autonomy.computer/docs/_for-coding-agents.md

Read all documentation thoroughly before writing any code.

Use the above documentation to create a new Autonomy app that news editors can use for fact-checking news stories.
The workflow is:
- The editor pastes an article into a text box.
- An agent then breaks the article into a set of claims.
- Each claim is then assigned to a parallel fact-checking agent.
- Fact-checking agents research the claim on the internet using the Brave Search MCP server.
- All the research is given to a report-generating agent that creates a final fact-check report.

Also, create a very simple web UI for the app using Next.js and shadcn/ui.
Before creating and deploying the app, create a README with an architecture diagram of the app.
Keep the app limited to one Autonomy node.

Deploy and test the app in Autonomy.
Unset NO_COLOR in your shell before running any autonomy command.
Your coding agent will:
  • Refer to the Autonomy documentation to learn how to create, deploy, and debug Autonomy apps.
  • Create a README describing the architecture of the application.
  • Write Python code for claim extraction, fact-checking, and reporter agents.
  • Write code to create parallel fact-checking deep-research agents.
  • Set up the Brave Search MCP server and create streaming APIs.
  • Build a Next.js user interface.
  • Deploy the application to the Autonomy Computer and test it.

​
Using Docker in Claude Cowork

Claude Cowork runs in a VZVirtualMachine (Apple’s Virtualization framework). To use Docker inside Claude Cowork, you need to expose your Docker daemon to the macOS VM. If you use Docker Desktop, enable “Expose daemon on tcp://localhost:2375 without TLS” in Docker Desktop settings, then create an SSH tunnel to expose it to the VM that Claude Cowork runs in. If you use Podman as your Docker daemon on macOS, you can expose it to Claude Cowork by setting up an SSH tunnel: On your Mac host (run this before starting Claude Cowork): 
# Start the Podman TCP service inside the Podman machine
podman machine ssh \
    "nohup podman system service --time=0 tcp://0.0.0.0:2375 > /dev/null 2>&1 &"

# Create an SSH tunnel to expose Podman to the host network
ssh -L 192.168.64.1:2375:localhost:2375 \
    -N \
    -p $(podman machine inspect --format '{{.SSHConfig.Port}}') \
    -i ~/.local/share/containers/podman/machine/machine \
    core@localhost &
Add the following to your prompt for Claude Cowork: 
Set the Docker host environment variable before running any Docker commands:
export DOCKER_HOST=tcp://192.168.64.1:2375

Install the Docker CLI binary if needed.
Verify Docker is accessible by running: docker ps

​
Learn More

Agents

Create deep work agents.

Tools

Give agents the ability to take actions.