Terms and conditions

Limn is made by Alex Wakeman, a sole trader in the United Kingdom, contactable at wakemana@gmail.com. These terms are governed by the laws of England and Wales, and the courts of England and Wales have exclusive jurisdiction.

Limn is software that runs on your own computer. You install and start it with npx limn-review. It reads pull-request activity from the GitHub repositories you choose, using your own GitHub login, and stores it in a file on your machine. We do not run a server for it and we receive none of that data.

This website has two parts: the pages describing Limn, and a contact form.

• Free for individuals, and for any organisation where no more than 3 people use Limn.
• More users than that (affiliates count together) need a commercial licence. Ask through the contact form.
• No competing product. You may not use Limn, or its source, to build or offer a product that competes with it.

This is a summary for convenience. The Limn License is the binding text.

Limn reads GitHub using your access token, so it can reach whatever you can reach. You are responsible for having the right to access every repository you add, and for following your employer's or client's rules about where that data may be copied and whether it may be sent to an AI model.

The AI features run on your own AI agent: a Claude Code login or Anthropic API key, or Codex on a ChatGPT plan. Content they send goes from your machine to Anthropic or OpenAI under your own agreement with that provider. We are not in that path.

On this website, you agree not to:

• probe, scan or overload the site (security research is welcome — see §8);
• use the contact form to send spam or anything unlawful.

AI-written reviews, fixes, notes and plans are suggestions produced by a language model. They can be confidently wrong: they miss real bugs, invent problems that do not exist, and misread intent. Nothing Limn generates replaces human review. A review is posted, and a fix pushed, only when you press the button or switch on the automatic setting yourself.

You are responsible for reviewing anything that is posted or pushed, and for anything you merge or act on. We accept no liability for a defect an AI review failed to find, or for a change made on the strength of one.

Only frontier models are supported for the AI coding tools (AI Review, AI Fix and the conflict resolver): the Claude models Limn offers, through Claude Code or an Anthropic API key, and the Codex models it offers, through a ChatGPT sign-in. Their safeguards and their judgement are part of what keeps a run safe. Pointing Limn at any other model is unsupported and at your own risk.

The AI features read code that other people wrote, and that code can try to instruct the model. We put a lot of effort into preventing data leaks, malicious prompt injection and unexpected calls to your computer:

• A path guard. Every file an agent reads or writes is checked against the pull request's own working copy. Paths outside it are refused.
• No shell for reviewers. A review or a chat can only read. The fixer can edit files but cannot run commands, so it builds and tests nothing.
• Allow-listed tools. Each agent gets a fixed list of tools. Anything else is refused before it runs.
• Sandboxing. Agents work in a throwaway copy of the repository, with no web access, and apart from your own Claude Code and Codex settings, plugins and connected accounts.
• Untrusted input. Anything an agent reads from a pull request is treated as untrusted, never as an instruction from you.
• Nothing pushed without you. Nothing is posted to GitHub or pushed until you press the button, unless you switch on an automatic setting yourself.

None of this is a guarantee. Models and the tools around them change, and new attacks appear. You remain responsible for what you run Limn on and for reviewing what it produces.

Limn and this website are provided as is, with no warranty of any kind, including merchantability, fitness for a particular purpose and non-infringement. GitHub is the system of record; everything Limn holds can be synced again from there.

You use Limn at your own risk, and all liability for that use rests with you. To the fullest extent the law allows, we are not liable for any loss, including indirect or consequential loss, lost profits, lost data, a leaked secret, a defect shipped or business interruption. Nothing here limits liability for death or personal injury caused by negligence, for fraud, or for anything else that cannot lawfully be limited — and if you are a consumer, your statutory rights are unaffected.

Found a security problem? Report it privately to wakemana@gmail.com rather than opening a public issue. We will not pursue good-faith testing that does not degrade the site or touch anyone else's data.

We will update the date at the top when these terms change. Every revision is in the public git history.