Many bug reports in open-source projects arrive as a screenshot and one line of text. Before anyone can fix them, a maintainer has to rewrite the report in the project's issue format and work out which file is responsible, and new contributors often get stuck at that step.
BugLens takes a public GitHub repository, a screenshot of the bug and one sentence. It returns:
- An issue draft that follows the repository's own issue template and uses only its real labels. Details nobody stated are marked [please confirm] instead of being invented.
- The top suspected files, each with a reason, a confidence value and the code excerpt that search matched, linked to the exact lines.
- A plain-language brief for a first-time contributor.
- Similar existing issues, so duplicates are noticed before filing.
Gemma 4 reads the screenshot, ranks the candidate files and writes the draft. Code search runs locally with an open-source embedding model and combines three signals: meaning, exact on-screen text, and file names seen in stack traces. The model can only choose among files that search found, so it cannot make up a path.
In our demo on a real, already-fixed bug from the File Browser project, BugLens ranked first the one file that the maintainers' fix changed, out of 353 files.
BugLens finds likely files. It does not fix the bug, and its output is a suggestion to verify before filing.
Top comments (0)