What is Maige and what does it do?

Maige is an open-source infrastructure project for running natural-language workflows on a codebase. It connects to a GitHub repository and lets you describe, in plain text, what should happen when issues and pull requests are opened — then it carries those actions out. Two conditions matter before you invest time: the hosted service has been sunset, and new signups are closed. The code remains public on GitHub as an archive, so Maige is now something to read, study, or self-host rather than a product to sign up for.

What Maige actually does

Maige describes itself as an "intelligent codebase copilot" and as "an AI with access to GitHub" that "can do anything you could do in the UI." In practice, its documented capabilities are:

• Auto-labelling — apply labels to issues and PRs based on rules you write.
• Auto-assignment — route issues to specific people (for example, UI-related issues to a named maintainer).
• Auto-comments — post comments on issues and PRs.
• Code review — review incoming pull requests against your contribution guidelines.
• Code generation and snippets — run simple code snippets inside a sandbox environment.
• Custom instructions — the whole behaviour set is configurable with text.

The rules are written as short natural-language statements. The site's own examples include assigning UI-related issues to a specific username, labelling .env PRs as needs-approval unless a maintainer opened them, and reviewing all incoming PRs per a CONTRIBUTING.md file.

How it works under the hood

When you connect a repository, Maige creates three things:

1. A webhook — so GitHub events reach Maige.
2. Embeddings of your entire codebase — so the system can reason over your code, not just the event text.
3. A sandbox environment — where it can execute simple code.

You then write your rules, and Maige runs them against incoming issues and PRs. A dashboard lets you monitor runs and give feedback. The project states it "works flexibly with the GitHub API to find a way," which is the mechanism behind the claim that it can do anything you could do in the GitHub UI.

A concrete example of the workflow: a contributor opens a PR that touches .env. Because you wrote a rule saying such PRs need approval unless a maintainer opened them, Maige labels the PR needs-approval and comments on it — no manual triage step. That is the pattern Maige is built around: express a policy in a sentence, let the system enforce it on every event.

Current status and availability

This is the most important part for anyone evaluating Maige today:

The site also lists a "version one: 2026/10/25" marker and says the project was used by 4.3k++ repositories, with logos from projects such as Documenso, Nuxt, Highlight.io, Cal.com, Trigger, and Precedent. Those figures describe past adoption, not current availability.

Pricing, as documented

The site lists a Standard Plan at $30.00 USD/month, with the first 30 issues free, covering auto-labelling, auto-assignment, auto-comments, custom instructions, code review, and code generation. An Enterprise tier is described as "best for large teams" and "coming soon." Because new signups are closed, treat these figures as historical documentation of the offer rather than something you can currently purchase. The page does not state whether self-hosting the open-source code carries any cost or licensing conditions beyond what is in the repository.

Who should look at Maige now

• Maintainers of open-source projects who want to see a worked example of natural-language GitHub automation — the rule syntax and the webhook/embeddings/sandbox architecture are the reusable ideas.
• Developers building similar tooling who want a reference implementation for connecting an LLM to repository events.
• Anyone who needs a working hosted bot today — Maige is not that, given the sunset and closed signups. You would need to self-host from the archive or choose another tool.

Maige was built as "an open-source experiment by Rubric Labs," and that framing is the right lens: it is best understood as a design and a codebase to learn from, not a service to adopt.