Was this pull request written with AI?

Paste any public GitHub pull request URL. We check it against open-source rules that describe the attribution each AI coding tool leaves behind — no account, nothing stored.

Public repositories only. Powered by the open-source detection rules Coderbuds runs in production.

What the detector looks for

Nothing here reads your code or guesses from style. Every rule matches an explicit marker that a tool wrote itself, which is why a result can be checked by hand:

Bot commit authors: The commit is authored or committed by the tool's own account, such as github-copilot[bot].
Commit message footers: A generated-by line the tool appends to the commit message or the pull request description.
Co-author trailers: A Co-authored-by trailer naming the assistant — the most common marker, and the one most teams do not realise they are publishing.
Branch naming patterns: Branch prefixes that a tool creates automatically rather than a human typing them.
Pull request labels: Labels applied by an integration to mark the change as agent-authored.

Tools with detection rules today

Aider
Claude Code
Cursor
Devin
GitHub Copilot
OpenAI Codex
Replit AI
v0.dev
WindSurf

Missing one? The rules are YAML — open a pull request.

What it cannot tell you

A negative result means no marker was found, not that a human wrote the code. Anyone who copies a suggestion into their editor and commits it themselves leaves nothing to match. Any team that strips trailers in a commit hook leaves nothing either.

That is a deliberate trade. A stylistic classifier would return an answer for every pull request and be wrong often enough that nobody could act on it. These rules return an answer you can verify by opening the commit — which is the only kind of answer worth putting in front of a team.

The reasoning, and how each tool was reverse-engineered, is in the write-up on open-sourcing these rules, and the case for measuring at the pull request rather than through vendor APIs is in why we track AI at the PR level.

One pull request is a curiosity. Every pull request is a number.

Checking a single pull request tells you what one developer did once. The questions engineering leaders are actually being asked need the whole set:

What share of our pull requests are AI-assisted, and is it going up?
Do AI-assisted changes reach production faster, or just wait longer in review?
Do they fail more often once they are out?
Are the seats we are paying for actually being used, and by whom?

Coderbuds runs these same rules across every pull request in your repositories, including private ones, and reads the result against your delivery metrics rather than on its own.

Track your team's AI adoption How the tracking works

Free trial, no credit card. Authorize your repositories and we sync the last 30 days of history, so there is something to read straight away.

Frequently asked questions

How does the AI code detector work?

It reads a public pull request from the GitHub API and matches it against open-source YAML rules that describe the attribution each AI coding tool leaves behind: bot commit authors, commit message footers, Co-authored-by trailers, branch naming patterns and pull request labels. Nothing is inferred from the code itself.

Does it work on private repositories?

No. The detector calls the public GitHub API without your credentials, so it can only read repositories that are already public. To analyse private repositories you need to connect them to Coderbuds, which runs the same rules using your own installation.

Can it detect AI code that has no marker?

No, and it does not guess. The rules only match explicit attribution. A developer who pastes AI-generated code into an editor and commits it by hand leaves no marker, so the detector will report the pull request as human-written. Treat a negative result as "no evidence", not as proof.

Do you store the pull requests I check?

No. The pull request is fetched, matched against the rules and the result is returned. Nothing about the pull request is written to a database.

Are the detection rules open source?

Yes. The rules are published as YAML at github.com/coderbuds/ai-detector under an open licence, so you can read exactly what each one matches, run them yourself, or open a pull request to add a tool.

You're subscribed!

Check your email for a confirmation link. You'll start receiving weekly engineering insights soon.

More like this, roughly weekly

What we learn measuring AI adoption, review load and delivery across real engineering teams. No product announcements.