An AI coding agent for your terminal.

Start with a focused repository question

Run LoopCode in the directory containing your project. The agent can inspect files, search the repository, edit code, and run commands subject to its permissions. An interactive session gives you a place to review requests, see tool output, and refine the task as the evidence changes.

cd /path/to/your-project
loopcode
# Or start with a request:
loopcode "Explain the test setup and find the checkout tests."

Resume the task instead of restating the repository

Use the conversation list to identify earlier work in this workspace. Continue the latest session with -c or resume a selected conversation with -r. Keep project requirements in versioned instructions and review any assumptions retained from an older conversation.

loopcode sessions
loopcode -c
loopcode -r <conversation-id>

Make automation output explicit

The -p flag runs a prompt without the interactive UI. Default output is the final answer on stdout, with tool progress on stderr. JSON is useful for a final result object; stream-json reports events for integrations that need progress.

A headless run has nobody to answer a new approval prompt. Actions not covered by the current permission policy are denied rather than silently approved. Begin with a read-only task and test the failure path before integrating a mutating run into CI.

loopcode -p "Summarize the public API. Do not modify files or run commands." \
  --output-format json --max-turns 12

Requirements and boundaries

• Node.js 22.18 or newer is required for the published CLI.
• macOS, Linux, and Windows have installation instructions; Windows uses a PowerShell installer.
• Hosted inference needs connectivity and an account with available plan usage.
• A terminal session is not the desktop browser integration. Use project tests and command output for terminal verification.