Already Filed
“Duplicate of #…”, without remembering thousands of issues.
Point it at a GitHub repo. An open model running in this tab reads every open issue (and the ones closed this month) and lists the open ones that say the same thing as an earlier one, with the comment ready to paste. Maintainers close duplicates in a click; reporters check before they file.
No server, no sign-up, no API key. GitHub sends the issues to your browser; the model never leaves it.
Hit GitHub's hourly limit? Add a token
Without one, GitHub allows 60 requests an hour from your network: a repo with 2,500 open issues uses about 30. A fine-grained token with no permissions at all (public repos only) is enough. It stays in this tab and goes only to api.github.com.
The reader is EmbeddingGemma, Google's open-weight embedding model. The first run downloads it, 197 MB, once.
Open issues that were probably filed before
0 of 0 open issues
Each one next to the earlier issues it most resembles, closest first. It suggests; you decide. Tested on two big repos it was never tuned on: the issue maintainers chose was among its top three for 65% of zed's duplicates and 58% of neovim's, and most of its other confident suggestions turned out to be duplicates nobody had linked.
0 more that might be (less sure)
About to file an issue?