Documentation

Install the GitHub App, add a policy file, your model key and a server, and follow one issue to a merged pull request.

The product, screen by screen, in the order work moves: Overview, Plans, Fleet, CI runs, Pending actions, Blockers and Audit log.

Add your own server to run the work: what it needs, the command that connects it, how to lock it down, and what never reaches it.

Use your own AI model key. Every model call runs on it, so your provider bills you directly.

The GitHub permissions the app asks for and why, how secrets are encrypted, and the audit log whose entries are never edited.

Every policy field, with examples. The file that bounds what the agent may do.

How we recognise a repository's stack, the three scripts it gets (install, run, check), and how a check says it could not test a change.

AI code review on your own model key. Keep your .coderabbit.yaml, or set review rules in your policy file.

Move review onto your own key without rewriting config: your .coderabbit.yaml keeps working.

How Dependabot and Renovate pull requests get reviewed, fixed and merged under your policy, plus a recommended dependabot.yml.

Remind, then close, issues and pull requests that wait on the person who opened them. Off until you set it, with the settings we recommend.

When the agent stops and asks a maintainer first: security, licences, breaking changes, and the other categories you switch on.

Test a policy change on past work: see which decisions a new maintainer.yml would change. Changes nothing, saves nothing.

Connect Claude Code, Cursor, ChatGPT or the Claude app to our MCP server, then file your first request and check the work.

Hand fixes to GitHub Copilot. Copilot opens the pull request; your policy decides whether it merges.

Send issues to Claude Code, running on your own machines with your own model key. Every call it makes to our MCP server is logged.

Send tasks to OpenHands, running on your own machine with your own model key. Every handoff is logged.

Run GitHub Actions jobs on servers you enrol by changing the runs-on line; public repos opt in first.

Signed events sent to a URL you own: what each one holds, how to check it came from us, retries, and which events exist today.

Send Sentry alerts to developerz.ai: each new error becomes a request the agent can turn into a fix, merged under your policy.

Files the agents make, like screenshots, reports and builds, kept in our bucket for up to 7 days, or in your own S3 bucket for as long as you choose.

We keep raw run, review and CI logs for 72 hours. Connect a Postgres database you own to keep them as long as you choose.

Build live boards from any MCP client or the REST API. They update on their own, and every widget only reads data.

Seats, spend caps and the Stripe billing portal, and why AI model costs never appear on our invoice.

Owner, admin, member and viewer: what each can do, how people are added, and the rules for changing roles.

The agent is quiet, the PR will not merge, the box is offline. The checks that resolve most reports.

Every tool on our MCP server: what it does and the permission it needs.

Reach your repos, tasks, policy, billing and audit log from your own code.

How many API reads and writes you get each minute, the limit on calls before your token is checked, and what an over-limit reply looks like.

AI dev, box, seat, lane, handoff, disposition. The words this documentation uses, defined once.