For IT, data and AI teams

Deploy and operate squads of AI agents in your own infrastructure, with every change supervised.

Agents work on branches and open pull requests. Your team reviews and merges. We install it in your environment with your team.

Talk to us How it works

How supervision works

Agents propose, people merge

Each agent works on its own branch and opens a pull request. Require a review in your git host's branch protection and nothing reaches your main branch until a person approves it.

Limits enforced at run time

Agents run in an OS sandbox by default: Seatbelt on macOS, bubblewrap on Linux. Credential stores such as ~/.ssh and ~/.aws are denied. Force pushes and git reset --hard are blocked.

Definitions and history in your repo

Squads and agents are markdown files in git. You change an agent the way you change code, with a pull request and a review.

A model for each agent

Each agent names its model and provider, on your own accounts. squads scoreboard reports, per model and role, how many runs succeeded and what they cost.

What you get today, and what you don't

Your security review will check this, so here it is.

Available now

  • Squads and agents defined as files in your git repo
  • Agent work delivered as branches and pull requests
  • Model and provider chosen per agent
  • Run history and a per-model scoreboard
  • OS sandbox by default, with credential stores denied
  • Open source CLI on your own machine

With our team

  • Installation on a server you control, done with your team

Not yet

  • Single sign-on (SAML, OIDC)
  • Role-based access control
  • Audit-log export to a SIEM
  • Published install recipes for AWS, Google Cloud and Cloudflare. We list a target only after a clean deploy on it.
  • Approvals from Slack
  • Compliance certifications. We hold none today.

Tell us what your review needs

We will go through your environment and your requirements, and say plainly what fits and what doesn't yet.

Talk to us