Vera FAQs

Question 1

What kind of output does Vera provide?

Accepted Answer

Vera generates token-efficient, structured output in JSON or markdown codeblocks. Each result includes crucial metadata like file path, line range, source content, symbol name, and type, making it easily consumable by AI agents and scripts.

Question 2

Does Vera require an internet connection or external dependencies to run?

Accepted Answer

No. Vera is a zero-dependency, single static binary with 60+ tree-sitter grammars compiled in. After a one-time `vera setup` to download curated ONNX models for embeddings and reranking, the entire search pipeline runs locally and offline, without external API calls or Python requirements.

Question 3

Can Vera utilize my GPU for faster performance?

Accepted Answer

Yes, Vera supports optional GPU acceleration. During `vera setup`, it can auto-detect your GPU (NVIDIA, AMD, Apple Silicon, Intel) and download the appropriate ONNX Runtime build to leverage your hardware for faster embedding generation and reranking.

Question 4

How does Vera achieve superior search relevance and accuracy?

Accepted Answer

Vera's unique retrieval pipeline uses cross-encoder reranking, which scores the query and candidate as a single pair for joint relevance. This method, combined with vector retrieval and BM25, significantly outperforms traditional methods, delivering proven high accuracy (e.g., 0.80 nDCG@10).

Question 5

What is Vera and what problem does it solve?

Accepted Answer

Vera is a high-accuracy, local-first code search tool built in Rust. It helps developers quickly find relevant code snippets by combining BM25 keyword matching, vector similarity, and advanced cross-encoder reranking across 60+ programming languages, all while providing structured, offline results.

Vera

Vera

Key Features

Use Cases

Key Features

Use Cases