Show HN: Claude Context but local – semantic code search without API keys(github.com)
github.com
Show HN: Claude Context but local – semantic code search without API keys
https://github.com/FarhanAliRaza/claude-context-local
1 comments
Looks really cool! As an Odoo developer (which means each customer's project is by definition a "large codebase", considering the size of Odoo itself) this could be really useful. Definitely am looking to try this out once my holiday ends.
The original requires OpenAI API keys for embeddings and Zilliz Cloud for vector storage. This version runs entirely on your machine using EmbeddingGemma and FAISS.
Key differences: - No API keys needed (uses local EmbeddingGemma model) - Your code never leaves your machine - Zero ongoing costs - Same semantic search quality
Technical details: - Tree-sitter for AST parsing to understand code structure - EmbeddingGemma (1.2GB) for semantic embeddings - FAISS for fast vector similarity search - MCP protocol for integration with Claude Code and other AI tools
Early benchmarks show ~70% reduction in token usage for Claude Code when searching large codebases.
Supports most major languages through Tree-sitter parsers - Python, JS/TS, Go, Java, JSX/TSX, Svelte, with more coming.
GitHub: https://github.com/FarhanAliRaza/claude-context-local
Would love feedback, especially on the approach to code chunking and embedding strategy!