I built an in-browser search engine that finds the exact function to fix a bug

I recently developed a project called Odin, which is an in-browser search engine designed to pinpoint the precise function needed to resolve a bug. The process is straightforward; you simply input a public repository URL and either link to an issue or describe the bug in your own words. Odin then downloads, parses, and ranks the functions according to the specified bug, offering a more targeted solution than traditional large language models (LLMs).

What sets Odin apart is its operational framework. It predominantly executes within the user's browser, employing Tree-sitter (utilizing WebAssembly) and small local models via WebGPU. This design ensures that code is not indiscriminately sent to a massive cloud LLM, thereby enhancing data privacy and security. The efficiency of Odin is noteworthy; when evaluated on 300 genuine issues from the SWE-bench Lite dataset, Odin successfully identified the correct file in the top 10 results 93% of the time, and the specific function 79.4% of the time.

Remarkably, it achieves this level of accuracy using models under 1 billion parameters, significantly reducing the cost per issue to a fraction of a cent.

Additionally, Odin offers a GitHub Action that can be integrated into your repository, automatically commenting on new issues with the most probable functions to remedy. I encourage anyone interested to give Odin a try, as your feedback, questions, and criticisms will be invaluable in refining its benchmarks. Odin is compatible with public repositories in Python, JavaScript, TypeScript, Go, Java, and Rust, making it accessible across a variety of programming languages. Simply visit allfather.dev to start using Odin for finding the code behind any bug report.