Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Launch HN: Greptile (YC W24) - RAG on codebases that actually works

Mar 05, 2024 - news.ycombinator.com
The co-founders of Greptile, a tool designed to answer questions about complex codebases, have introduced their product on Hacker News. Greptile uses Language Model Learning (LLM) to comprehend entire codebases and answer difficult questions with full context. The tool parses the Abstract Syntax Tree (AST) of the codebase, generates docstrings for each node, and embeds the docstrings. It also uses vector similarity search, keyword search, and an "agentic search" where an agent reviews the relevance of the search results and scans the source code for important references.

The tool is currently being used by developers at Stripe and Microsoft for debugging, understanding open-source repositories, and parsing legacy code. Greptile ensures security by not storing any code on their servers after initial processing and only pulling snippets as needed from the GitHub API. The tool can be tried on popular open-source repositories or on personal repositories. The founders are seeking feedback to improve the product and are interested in hearing experiences of working with complex codebases.

Key takeaways:

  • Greptile is a tool that helps developers understand complex codebases by accurately answering questions about them, allowing developers to spend less time wrestling with codebases and more time writing code.
  • Greptile uses a method called "agentic search", where an agent reviews the relevance of the search results, and scans the source code to follow references that might lead to something important, providing more comprehensive and accurate results than simple RAG.
  • Developers at companies like Stripe and Microsoft are using Greptile for debugging, understanding open source repos, and parsing legacy code, especially when original engineers have left the company.
  • Greptile takes security seriously, not storing any code on their servers after initial processing and only pulling snippets as needed from the GitHub API. Their permissions are read-only and the only thing they do "on your behalf" is read code, so they can index the repo.
View Full Article

Comments (0)

Be the first to comment!