Duplicate of #...: an open model that remembers 23,000 GitHub issues so maintainers don't have to

Description and use

The "Already Filed" web app automatically detects duplicate GitHub issues in large repositories using embeddings and keyword search, presenting a copy‑ready duplicate comment to reduce manual search time.

Problem

Maintainers and contributors waste time searching for duplicate issues in large repositories, often missing older duplicates because they are not remembered.

Outcome

In zed, 65% of duplicate links were among the top three suggestions; in neovim 58%. The combined embedding and keyword recipe outperformed keyword search alone (38% top‑3). The app runs entirely in the browser, requires no API key, and processes 2,700 issues in about five minutes on a laptop.

Workflow

• 1. Point the app at a public GitHub repo or use the pre‑loaded ollama snapshot.
• 2. The app fetches all open issues and those closed in the last 30 days via GitHub’s public API.
• 3. Issues are cleaned, embedded with EmbeddingGemma (or a smaller model), and scored using a weighted mix of body, title, and keyword vectors.
• 4. The app lists open issues that resemble earlier ones, showing the most similar earlier issue and a button to copy a duplicate comment.
• 5. Maintainers or reporters review the suggestions and decide whether to mark the issue as duplicate or file a new one.

Evidence

In zed-industries/zed, maintainers have written it 1,653 times, and the issue they pointed to was filed a median of 1,343 issues (64 days) earlier.

Source

Point it at a GitHub repo and an open model, running in your browser tab, reads every open issue (and every one closed in the last 30 days), then lists the open issues that say the same thing as an earlier one.

Source

On zed’s 1,653 duplicates, the issue a maintainer pointed to is in the top three suggestions 65% of the time, against 38% for keyword search.

Source