Installation

baro ships as an npm package called baro-ai. The published artifact bundles the CLI binary (Rust) and the TypeScript orchestrator together — one install gets you both.

Prerequisites

• Node 20+ for the npm install path.
• An LLM backend. baro runs end-to-end on any of these — pick one, or set up several and switch with --llm per run:
  • Claude Code CLI for --llm claude (the default). baro shells out to the Claude Code CLI and bills against your Claude subscription. See claude.com/code.
  • Codex CLI for --llm codex (and the hybrid preset). Bills against your ChatGPT Plus/Pro subscription.
  • OPENAI_API_KEY for --llm openai. baro calls the OpenAI API natively — no CLI in between. Works with any OpenAI-compatible endpoint too (--openai-base-url — OpenRouter, vLLM, Ollama, DeepSeek…).
• gh (optional) if you want baro to auto-open the pull request when a run finishes. Without it, baro pushes the branch and stops.

See LLM providers for the full comparison — what each path does under the hood, model defaults per phase, and when to pick which.

Install

npm install -g baro-ai

Verify:

baro --version

You should get baro 0.82.x or newer. baro checks for new releases on start and prints a notice when your install is behind.

Verify your environment

Run the built-in self-diagnostic:

baro --doctor

You'll get a pass/fail report for each check baro depends on. Every check should print ✓. If any of them fails, the report tells you specifically what's wrong and where to look — see Troubleshooting for the longer walkthroughs.

Your first run

Once baro --doctor is fully green, give baro a goal:

cd path/to/your/repo
baro "Add a CONTRIBUTING.md at the repo root"

See Quickstart for the end-to-end shape of a run.