Breakwatch

APIs change. Your code shouldn't break.

When an API you depend on changes or retires something, Breakwatch finds the affected code and opens a pull request with the fix. Nobody reads changelogs. You merge a PR.

Starting with OpenAI and Anthropic: models get switched off on a date, and the code still calling them breaks.

Or get early access to the GitHub App that does it for you.

17OpenAI models shut down on 23 Oct 2026 alone.
13 of 50active repos we sampled, whose code names a retired model, still send requests to it.
2 mergedFirst fixes merged by neuron-os and text-generator.io.

How it works

  1. Finds. Every model name in your code, checked against OpenAI's and Anthropic's own deprecation pages.
  2. Fixes. Picks the replacement and changes what it rejects: newer Claude models refuse temperature, newer OpenAI models need reasoning_effort and max_completion_tokens.
  3. Proves. Runs your tests, then opens the PR. You merge it or you don't.

Why not just swap the name?

// before
model: "claude-3-5-sonnet-20241022", temperature: 0.2   → 404, the model is gone
// swap the name only
model: "claude-sonnet-5-5", temperature: 0.2            → 400, it rejects temperature
// Breakwatch
model: "claude-sonnet-5-5"                              → works

Get early access

The GitHub App watches your repos and opens the fix PRs before an API change breaks your code.

Thanks, you're on the list.