Bulava

Stop babysitting AI.
Claude Code writes, Codex reviews, and neither of them asks you to press continue.

Download for macOS · Install the CLI · How it works · The benchmark

---

Bulava runs Claude Code and the Codex CLI on your Mac, on the subscriptions you already pay for. You describe the work; Claude does it, Codex reviews it independently and sends it back until it is actually done. There is no account to create, no API key to paste, and no server of ours in the path: everything happens on your machine, in your folders.

It started as a shell script whose whole purpose was to stop a model announcing "I have done 30% of the work — shall we leave it here?" and then asking a human to say no. Ninety-odd days later it is the application that writes itself.

One window. Products on the left, the agent's real transcript in the middle, and what you can act on — changes, evidence, reports — on the right.

One message, and everything that happens to it

This is the arrangement the release on [bulava.app](https://bulava.app) runs. Every stage below is traceable to a file: the stage list to `engine/supervisor/pipelines/adaptive-peer.json`, the branch at the top to `engine/bin/preflight.sh`, the numbers on the review loop to `engine/supervisor/config.sh`.

flowchart TD
    MSG([Your message — typed, dictated or dropped in