Free · macOS and Linux · Source-available

Pick your team. Hand off the work.

Describe what you want built. brindle runs your AI coding agents as an autonomous engineering team that plans, builds and reviews the work for you. You come back to finished work, without any of the babysitting or the decision fatigue.

Works with Claude Code, Codex, Google Antigravity, OpenCode and open-weight models

Checked, not claimed.

brindle treats each agent's work like a transaction. Your configured checks pass before it merges, with approval of that exact commit when review is enabled. Otherwise the merge is refused.

Done means a command passed.A milestone is done when its check command exits 0, and brindle re-runs the checks after later merges.
A reviewer checks the work.With review enabled, a reviewer approves the exact commit before it merges. You can choose Codex or a local model to review Claude's branches.
Runs on your machine.On paid plans, enabled analysis features send task text and file paths for review briefs and upload documents that can quote code. Client switches start off unless you or your org enable them. Admin-published profiles and rule packs also reach PawDelta. Details are in Privacy.

brindle itself and this website are developed in brindle sessions, with their own checks.

Why brindle?

Because the work should run without you standing over it.

  • Describe it once and get on with your day.

    The supervisor splits the goal into tasks, sends each to a model sized for it, and keeps going until every milestone's check passes or it needs you.

  • Come back to work that's already reviewed and tested.

    With review enabled, a reviewer reads every branch and sends weak work back to its worker before it ever reaches you. Nothing merges until your configured checks pass.

  • Your agents, your models, your machine.

    Mix Claude Code, Codex, Google Antigravity, OpenCode and open-weight models, with a local model for workers and reviewers if you like. You choose the providers that handle your code.

  • Say it once. It sticks.

    Record a decision and every session on the repo follows it. Took a wrong turn? Rewind a worker to an earlier turn and set it going again with your note.

How brindle compares with other tools

Plan as a team. Build with brindle.

Sign off once. Agents build in parallel.

A shared plan doc for adding CSV export to reports, with comments from product, design and support, a tracked limit suggestion from engineering, and reviewer sign-offsA shared plan doc for adding CSV export to reports, with comments from product, design and support, a tracked limit suggestion from engineering, and reviewer sign-offs

See how team planning works →

Watch one goal run

One message from you, start to finish, in the brindle desktop app for macOS and Linux, simulated with the same steps a real session takes.

Simulated8/8One goal, start to finish

1 message from you · 4 tasks · 4 branches merged · 1 bug caught before merge. brindle planned, built, reviewed, fixed, checked and merged the rest.

Start with the desktop app

Free, on your machine. The app needs a free account, with no payment. Download it, open it, sign in, and point it at a repository when you are ready.

  1. Download the app. For macOS or Linux, from the downloads page. It drives an agent CLI you already use, such as Claude Code or Codex.
  2. Watch it run first. From New session, you can choose a short demo on a practice repo, so you can watch a goal run before it touches anything of yours.
  3. Open your own repo. Point the app at the folder and hand it a goal. Getting started covers the checks a branch must pass.

brindle is free to use, on your own machine and at work. Source-available under the Brindle License 1.0: you can read the source, write plugins and make private fixes, but it isn't open source (no redistribution or other modification without permission). See the terms.