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!