Assign it an issue.
Get back a merge request.
Mate joins your GitLab as a real user. It works in an isolated sandbox with only the permissions you grant, uses your own model keys, and hands you a merge request to review.
- ✓ GitLab.com & self-managed
- ✓ EU-hosted or on-prem
- ✓ No tracking
From a simple instruction to a single task or a whole app, you're in charge.
From issue to merge request in four steps
- 01
Assign or mention
Mate is a real GitLab user. Assign it an issue or a merge request, request its review, or @mention it in a thread. It also reacts to failed pipelines.
- 02
Nothing inside to leak
Each job gets a fresh container built from your devcontainer, with no GitLab or model credentials inside to leak. It still acts in GitLab, through the gate.
- 03
Only what you allow
Every GitLab action is checked against the permissions you gave that agent. It never approves its own work unless you choose to.
- 04
You stay in charge
Comment on the merge request and mention Mate: it pushes follow-ups to the same branch and answers in the thread. Or take over and push yourself.
Your model, your keys
Claude Code, Codex, OpenCode or omp, with Anthropic, OpenRouter or any OpenAI-compatible endpoint. No token markup.
Safe on untrusted code
Build steps run in a locked-down sandbox with no secrets. Your model key never enters the agent's container.
EU-hosted, or yours
Hosted in the EU with no tracking cookies, or entirely on your own servers.
Why not just…
…paste code into a chat?
A chat has never seen your repository, your CI or your review rules. Mate starts from the issue, reads the project, and answers in GitLab where your team already works.
…use an IDE agent?
It lives in one developer's editor and session. Mate works in the background on its own GitLab account, in a sandbox, while the team gets on with other work.
…script it ourselves?
You can. Then you own webhooks, per-job credentials, sandboxing, queueing and a permission model. Mate is that work, already done and maintained.
Pay per teammate, not per developer
Isolated container per job on shared EU workers.
Your own worker that sleeps when idle.
Unlimited mates on your own infrastructure.
Pay yearly in advance by bank transfer, or monthly by card (coming soon) (10 times the monthly price, a proforma invoice by email). Final price, no VAT charged (we are not a VAT payer). Model usage is billed by your own provider, at their price. Full pricing and FAQ
Built by one engineer, not a data company
I'm Ondřej Beňuš. Mate is my own project: no investors, no ads, no collecting your data. I build it because I want an AI teammate I would trust with my own code. My name is on it, and so is my reputation.
Read the story →Questions, answered
- Does it only work with GitLab?
- Yes. Mate is built around GitLab webhooks, bot accounts and permissions, on GitLab.com and on self-managed instances. There is no GitHub support.
- What can Mate do in my projects?
- Only what you grant each agent:
read,comment,push_branch,open_mr,merge,close_issue,update_issue,touch_ciand the opt-inapprove_mr. Every GitLab call goes through a permission gate that checks it again on the server. It never approves its own work unless you choose to, and GitLab's own approval rules still apply. - What starts a job?
- Assigning Mate to an issue or merge request, requesting its review, mentioning it in a comment, or a failed pipeline on a branch you choose. Labels can narrow when a trigger fires. You can combine these with several agents, each with its own permissions.
- Which models and providers can I use?
- Four agent CLIs: Claude Code, Codex, OpenCode and omp. Bring a key from Anthropic, OpenRouter or any OpenAI-compatible endpoint. Mate adds no markup and does not bill for tokens. You pay your provider directly.
- Where does my code go?
- The repository is cloned into a short-lived container that is destroyed after the job. Code and prompts go from that container to the model provider you chose, through a sidecar that holds your key. Job events and logs are kept for 30 days so you can follow a run in the console. With on-prem, none of it leaves your network.
- Can I run it on my own servers?
- Yes, under a yearly contract: signed container images plus an install guide, running next to your own Postgres. It is the same product as the hosted one. Write to support@mate.engineer.
Try it on a real issue this week
Register, connect your GitLab and order your first mate. Setup takes a few minutes.
Already have an account? Sign in