We’re excited to share bloop, a new way to understand your codebase. We’ve combined three bleeding edge technologies into a lightweight desktop client:
- A blazing fast code search and regex matching engine written in Rust
- Precise code navigation, built on stack graphs and scope queries
- An intelligent AI agent, powered by GPT-4 and semantic code search
Like many developer tools, bloop was born out of necessity. It’s early 2021 and we’re frustrated with the existing options for code search. Our friends at GAFA tell us about internal experiments, and research papers hint at a future where machine learning applied to code is helping 10X developers become 100X.
So we built a prototype semantic code search engine, turning a user’s query and each file into vectors and comparing them for similarity. It was a step in the right direction, but the query parsing was basic so that “package.json” would match similar results to “everything except package.json” and we could only retrieve code, we couldn’t explain, summarise or update it.
Large Language Models proved a turning point. Rather than just return chunks of code, these incredibly sophisticated networks could understand code at a higher level, and reason about it. All of a sudden we seemed much closer. Our own use of bloop went up.
We’re extremely proud of what we’ve achieved a few iterations later. We’re standing on the shoulders of Open Source giants like Tantivy, Tauri and Qdrant and are pleased to be fully Open Source ourselves.
Copilot has transformed the way we write code, we hope that bloop will change the way you read it. You can get started here.