Goal

Component to build and query fast, pure keyword search index (BM25/TF-IDF/inverted index) — useful for retrieval steps of RAG pipelines. State is just index bytes; host controls persistence.

Scope/Features

• Build index: documents in, index bytes out
• Query index: index bytes in, text query in, top-K hits out
• Stateless, pure; no file/network access

Rough WIT shape

• build(docs: list<doc>) -> list<u8>
• query(index: list<u8>, q: string, k: u32) -> list<hit>
• Doc/hit typedecls.

Candidate crates

• tangopool/tantivy
• noodle

Labels

• help wanted

---

Suggest discussing v1 (pure/byte model) and possible evolution toward incremental/resource-based interface for scaling.