Rawq – semantic code search for AI agents (4x fewer wasted tokens, Rust, OSS) (github.com)

Rawq, a new semantic code search engine tailored for AI agents, has been announced, boasting a significant reduction in token usage—four times fewer wasted tokens when processing code. Developed as an open-source Rust binary, Rawq is designed for fully offline operation, making it efficient and practical for searching vast codebases. Rather than returning entire files, Rawq intelligently retrieves only relevant code snippets, along with file paths, line ranges, and confidence scores. This capability is particularly crucial for AI agents that consume tokens while analyzing code, as it enhances the accuracy of their output and reduces unnecessary data processing. Significant for the AI and machine learning community, Rawq employs a hybrid search approach that combines semantic (using ONNX embeddings) and lexical (BM25 algorithm) strategies, yielding effective results across 16 programming languages. It supports advanced features like GPU acceleration and daemon mode for optimized search operations, and allows for incremental indexing and efficient change detection, thus improving developer productivity. With its structured output format and user-friendly terminal experience, Rawq not only streamlines code searches but also empowers AI agents with precise context, fostering improved integration of AI into development workflows.