Show HN: VectorDB-CLI – Semantic code search powered by local embedding models

1 point

a year ago

I built a command-line tool that lets you search your codebases semantically rather than just with regex or string matching. It uses vector embeddings to understand the meaning behind your code and queries.

*Features:* - Indexes multiple Git repositories and tracks specific branches - Syntax-aware parsing for 8 languages (Rust, Python, JavaScript, TypeScript, Ruby, Go, YAML, Markdown) - Uses ONNX embedding models running locally (no API dependency) - GPU acceleration support on Linux (CUDA) and Mac (Metal) - Search with natural language queries like "where is the authentication logic?"

I developed this over 2½ weeks with Cursor, which produced about 12,700 lines of Rust code with Claude 3.7 and Gemini 2.5 as my LLMs of choice. Each has different strengths - Gemini is great for quick specific tasks, while Claude excels at detailed implementations requiring attention to detail.

Currently working on adding a gRPC layer to make it more accessible for integration from any language, enabling custom coding agents or other AI tools. It's MIT licensed and freely available as a crate at https://crates.io/crates/vectordb-cli with repository at https://gitlab.com/amulvany/vectordb-cli.

Building this has been quite the learning experience with many mistakes and rollbacks, but it's now becoming genuinely useful for navigating complex codebases.

Would appreciate feedback from the HN community on what capabilities you'd like to see added!