Your backlog, shipped overnight.
File a GitHub issue. Come back to a reviewed pull request. mago runs a small AI engineering team inside your repositories — reading the issue, writing the code, reviewing it, and waiting for your approval to merge.
Runs on your machine with your LLM key. Your code never passes through us, and we never resell tokens.
Real pull requests, written by mago
Every one of these was opened, reviewed and merged in a public repository (javimosch/memgraph) from a GitHub issue, by an agent reading the codebase. Nobody wrote the code by hand.
Fixed a silent data-routing bug
A registry entry recorded the wrong scope, so writes landed in an unrelated project. It found the cause, fixed both code paths, and added warnings for existing bad entries.
+366 −54Turned warning spam into a repair command
Nine warnings on every single command. It narrowed them to real problems and added
projects --repair — refusing to auto-fix the ambiguous case.
Fixed an argument parser
--version exited with an error code. So did any flag placed before the
command. One fix closed two separate issues.
It also cut the release — version bump, changelog, tag, published binary and checksum.
Three things you do. That's the whole job.
Describe the work
Open a GitHub issue like you would for a teammate. One line is enough.
Add the label mago.
Let it work
An agent picks it up, reads your codebase, writes the change, and opens a pull request. A second agent reviews it against your tests.
Review and merge
You get a normal PR with a normal diff. Comment on it, ask for changes, or merge it. Nothing reaches your main branch without your say-so.
What people actually point it at
The bug backlog nobody gets to
Small, well-understood bugs that have sat open for months because they are never the most important thing. This is where it earns its keep.
Test coverage
"Cover the error branches in this file." Tedious, mechanical, valuable, and easy to verify — the shape of task it is best at.
Papercuts
The confusing error message. The flag that does not work. The thing everyone works around and nobody schedules.
Docs that drifted
READMEs describing a version you shipped two releases ago, command tables missing half the commands.
Release chores
Version bump, changelog, tag, publish. It can run the whole sequence and stop if the build fails.
Small, specified features
Where you can write the acceptance criteria in the issue. If you cannot describe it, it will ask rather than guess.
Get started in about five minutes
What you need
- A GitHub repository you can install an app on
gitand the GitHub CLI, signed in- An LLM you already pay for — a Claude subscription, or any provider API key
You do not need to "use AI agents" already. If you can copy two commands into a
terminal, you can run this. Prefer to stay in the terminal entirely?
mago register creates the account without the browser.
- Create your account with GitHub — 48 hours free, no card. You get a short setup code.
-
Install the CLI
curl -fsSL https://mago.intrane.fr/install.sh | sh -
Attach this machine to your account, using the code from step 1
mago claim MG-XXXX-XXXX -
Connect your repositories, then start the worker
mago link && mago serve --relay -
Label any issue
mago— and watch the pull request arrive
Stuck? mago worker doctor checks every prerequisite and tells you exactly
what is missing.
The team you get
Four roles, each a plain markdown file you can read and edit. They share notes on what they learn about your codebase, so the work gets cheaper and sharper over time.
CTO writes the code
Picks up engineering tasks and ships them as tested pull requests. Never merges its own work — that separation is enforced, not suggested.
Head of Org Engineering reviews
Reviews every pull request and runs your build and tests before approving.
Head of Product plans
Turns a vague issue into a concrete spec, and asks you when something is genuinely ambiguous instead of guessing.
CMO writes
Release notes, README updates, announcements — the writing that never gets done.
The questions people actually ask
Can it merge to my main branch without me?
Not by default. Out of the box it opens a pull request, reviews it, and stops — you merge. If you later want it to merge on its own, you can turn that on, and even then it only merges when your own build and tests pass.
What if the code it writes is wrong?
Then you close the pull request, the same as with any contributor. You are reading a normal diff in GitHub before anything lands. A second agent reviews first, and your test suite gates it.
Does my source code get sent to you?
No. The worker runs on your machine and talks to your LLM provider with your key. We relay GitHub events to it and handle billing. We never see your code, and we never resell tokens.
What does it cost to run?
€20/month for mago, plus whatever your LLM provider charges for the work it does. When nothing is labelled, nothing runs, and the cost is effectively zero.
Do I have to work from the terminal?
Only to install it and start the worker — about three commands, once. After that you live in GitHub: issues in, pull requests out.
What if I want a dashboard instead?
mago is deliberately GitHub-native, with no panel to babysit. If you would rather have a managed UI, Automaintainer is the UI-driven alternative from the same workshop.
One plan. No tiers.
- 48-hour free trial — no card
- Unlimited repositories and projects
- The whole team, every role
- Bring your own LLM key — we never charge per token
- Cancel from the terminal, any time