v2.0.0

What's new

• Index format v3 keeps embeddings and source byte ranges; search reads matching text from the original files and verifies that those files have not changed.
• rag keyword searches live files with ripgrep-style patterns and ignore rules, including every source in rag.toml by default.
• rag search --hybrid combines semantic and keyword ranking. rag.toml can set the default and weights, and --only selects named sources for federated search.

Upgrade

Version 2 cannot read v2 indexes. Rebuild them with rag index in a directory containing rag.toml, or use rag index PATH -o INDEX_DIR for a standalone index. Keep the source files available when searching; v3 indexes do not contain chunk text.