Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep

Semble released an open-source code search system for coding agents that blends static embeddings, BM25, fusion ranking, and reranking.

The tool is designed to index and query large repositories on CPU, aiming to help agents find relevant code faster and with fewer tokens.

Semble says it uses 98% fewer tokens than grep-plus-read workflows while matching near-transformer retrieval quality on its benchmark.

The project targets agentic coding tools like Claude Code, Cursor, Codex, and OpenCode, and is available as an MCP server.