Skip to content

Mate documentation

Assign an issue. Mate does the work.

Mate is a GitLab-native agent orchestrator. You add a bot user to your project, assign it to an issue or MR, mention it in a comment, or configure it to react to labels and pipeline failures — and Mate spawns an LLM-driven agent in an isolated container. Today the agent reads the context and reports back as GitLab comments; broader write actions (branches, MRs, issue updates) are part of the permission model but their tools are not yet enabled during the closed beta. It never has access to your credentials and leaves no side-effects outside of GitLab.

Mate is currently in closed beta. A limited number of tenants are admitted from the waitlist as capacity becomes available. If you do not have an invitation, sign up at platform.mate.engineer and you will be placed in the queue.

Set up your first mate

A step-by-step walkthrough: create a tenant, connect GitLab, add an LLM key, register a webhook, define an agent, and trigger your first job.

Start the walkthrough →

Reference

Concepts, trigger configuration, permissions, models, security, pricing, and troubleshooting.

Browse the reference →

  • A bot user lives inside your GitLab project. It is a real GitLab account — every action the agent takes appears under that user’s name.
  • You trigger the agent by assigning the bot to an issue or MR, mentioning it in a comment, applying a label, or letting it watch for pipeline failures. Triggers are configured per-project in .mate.yml or through the console.
  • Each trigger spawns a job: an isolated container with the agent runtime, a short-lived clone token, and scoped GitLab permissions. The container is destroyed after the job finishes.
  • The agent reports back as GitLab comments (with merge request and branch actions arriving as the beta tool surface expands). It cannot approve MRs — that separation of duties is intentional.

These docs cover the hosted Mate service (running on platform.mate.engineer). The on-prem edition — for running Mate on your own hardware — has separate installation and operations documentation that is delivered together with the on-prem release.