Getting Started

Install DocsClaw and deploy your first AI agent in minutes.

Install

Homebrew

brew install pavelanni/tap/docsclaw

Go Install

go install github.com/redhat-et/docsclaw/cmd/docsclaw@latest

Container

podman run --rm ghcr.io/redhat-et/docsclaw:latest version

From Source

git clone https://github.com/redhat-et/docsclaw.git
cd docsclaw && make build

Pre-built binaries for Linux, macOS, and Windows (amd64/arm64) are available on the releases page.

Agent Configuration

An agent is defined by a config directory with three files:

my-agent/
  system-prompt.txt — agent personality (required)
  agent-card.json — A2A discovery metadata (optional)
  agent-config.yaml — tools and loop settings (optional)

The system prompt defines what the agent does. The agent card tells other agents how to discover and interact with it. The agent config enables tool use (shell commands, web fetch, file operations) and controls the agentic loop.

Example agent-config.yaml for a tool-enabled agent:

tools:

  allowed:

    - exec

    - web_fetch

    - read_file

    - write_file

  exec:

    timeout: 30

    max_output: 50000

  workspace: /tmp/agent-workspace

loop:

  max_iterations: 10

Run the Agent

Set your LLM API key and start the server:

      # Using OpenAI

      $ export LLM_API_KEY=sk-...

      $ docsclaw serve --config-dir my-agent/ --listen-plain-http

      INFO Agent name=my-agent

      INFO Listening addr=0.0.0.0:8000

      # Using Anthropic

      $ export LLM_PROVIDER=anthropic

      $ export LLM_API_KEY=sk-ant-...

      $ docsclaw serve --config-dir my-agent/ --listen-plain-http

Talk to the Agent

Use the built-in interactive chat TUI:

$ docsclaw chat --agent-url http://localhost:8000/a2a

Or use the A2A CLI for scripted interactions:

      $ a2a discover http://localhost:8000

      $ a2a send http://localhost:8000 "Summarize https://go.dev/blog/go1.24"

Next Steps

Explore the demo scenarios for ready-made agent configurations, or check the GitHub repository for the full documentation.