Measure your RAG.
Or keep guessing.

What we measured

On our own eval corpus: fixed-size chunking breaks mid-sentence 83.3% of the time; recursive chunking, 39.6%. That's the kind of number you want before you ship a RAG system — not after.

What's inside

• Six chunking strategies — fixed, recursive, semantic, document-based, propositional, agentic
• Three retrieval modes — dense baseline, hybrid with reranking, agentic with an LLM judge
• Contextual retrieval & HyDE for harder queries
• RAGAS-style eval harness — faithfulness, answer relevancy, context precision/recall
• Query from the CLI or straight from your editor via MCP

Local-first. Your notes never leave your machine.