TLDR: I built a search engine that reads a bug report (or lets you define it) and points to the exact file (93%) and to the exact function (79.4%) you need to fix.

Hey everyone,

I recently made a new project while trying to get into RAG and automation.

Odin is more than a search engine, all you have to do is paste a public repo URL and link the issue or describe the bug yourself and it downloads, parses and ranks the functions based on the mentioned bug.

Why its different than any LLM:

It runs mostly in your browser: It uses Tree-sitter (via WebAssembly) and small local models (via WebGPU) so your code isn't just blindly shipped off to a giant cloud LLM.

It punches above its weight: Evaluated on 300 real issues from the SWE-bench Lite dataset, it puts the correct file in the top 10 results 93.0% of the time, and the exact function 79.4% of the time.

It’s incredibly efficient: It matches the accuracy of heavy Claude 4.5 agents using models under 1B parameters, cutting the cost per issue to a fraction of a cent.

There’s also a GitHub Action you can drop into your repo to automatically comment on new issues with the most likely functions to fix.

I would really appreciate it if you would give it a try (it would help me with the benchmarks as well)

feedback, questions, criticism are welcome