# Vera

A local code search tool for developers and AI agents that combines BM25 keyword search, vector similarity and cross-encoder reranking, returning symbol-bounded results instead of whole files.

Vera indexes a codebase with tree-sitter parsers for over 60 languages and searches it with a hybrid of BM25 and vector embeddings, optionally reranked by a cross-encoder, returning chunks bounded to a symbol's definition along with file paths, line ranges and scores rather than entire files. It ships as a single static binary with no per-language toolchain to install, and can run entirely offline on CPU with a local embedding model or use a hosted embedding and reranking service such as OpenRouter, OpenAI, Jina or Voyage for higher search quality.

The README reports a blind-graded benchmark in which the hosted embedding-and-reranker configuration used roughly half the prompt context of giving an agent no search tool at all, at equal answer quality, and describes incremental index updates and a watch mode that keep the index current as files change.

It is written in Rust.

Website: https://github.com/VeraTools/vera