Headless & remote execution.

What is headless mode?

Headless mode runs the Rabbit agent without any GUI. It uses the same agent loop, tools, and rules as the IDE but operates entirely via stdin/stdout or over SSH.

Remote via SSH

# Install once on the remote box, then run a task non-interactively
ssh dev-server "rabbit -p 'optimize the database queries in src/db/'"

# Provider + key come from the remote environment
ssh dev-server "ANTHROPIC_API_KEY=… rabbit -p 'add tests for src/auth.ts'"

Background agents

The IDE supports spawning background agents that run in isolated git branches. Each background agent:

• Creates a working branch automatically
• Runs up to 3 concurrent agents
• Respects budget caps per agent
• Can auto-create a pull/merge request on completion

Docker / devcontainer

Rabbit IDE supports VS Code devcontainers. The agent operates inside the container, ensuring tools run in the correct environment. Shell commands execute in the container, not on the host.

Scripting the agent

For automation -- Slack bots, internal dashboards, scheduled jobs -- shell out to rabbit -p "<task>" and capture its streamed output. Each invocation runs one task against the configured provider and exits, which keeps it easy to wrap from any language.