Show HN: Vexus – tells Claude Code where to look before it greps your repo (github.com)

Vexus, a new tool for developers, has been announced to enhance the capabilities of Claude Code by optimizing how it searches code repositories. By utilizing a hook that runs with every prompt, Vexus proactively locates relevant code snippets in a repo before the agent executes its searches. This approach reduces the number of tokens required during the interaction—achieving up to 10% cost savings in live Claude Code sessions—while maintaining efficiency and accuracy in locating critical code elements. The tool indexes locally, requiring no additional learning or tool definitions and effectively streamlining the process of finding code references. The significance of Vexus for the AI/ML community lies in its innovative method of integrating local search capabilities with a generative model, enabling more sophisticated understanding and retrieval of coding patterns and functions. This hybrid model of retrieval combines vector-based and keyword search techniques, allowing for quick querying and semantic context awareness while minimizing redundant API calls. With support for multiple programming languages and a structured indexing mechanism, Vexus is positioned to significantly improve developer productivity by addressing the limitations of traditional full-text search methods employed by agents like Claude Code.