Show HN: Save Claude tokens with semantic search powered by SQLite and Ollama (github.com)

The launch of Lumen, a fully local semantic code search engine for AI coding agents, marks a significant advancement in the AI/ML community's ability to enhance coding efficiency. By utilizing open-source embedding models, SQLite, and local computing resources, Lumen optimizes the process by allowing AI (like Claude) to locate relevant code snippets based on meaning rather than reading entire files. This approach drastically reduces costs—demonstrated by benchmark tests showing up to a 39% cost reduction and a 66% decrease in output tokens across various programming languages, all while maintaining patch quality. Key technical developments include Lumen's innovative use of Merkle tree change detection and semantic chunking, which enables fast re-indexing and efficient storage of large codebases. It supports a wide range of programming languages and minimizes dependency on external databases or APIs, ensuring data privacy. With reproducible benchmarking results and a user-friendly setup that requires only a static binary and a local embedding server, Lumen empowers developers to leverage AI functionality without sacrificing performance or incurring additional costs.