Changelog¶
Full changelog is maintained in CHANGELOG.md in the repository root.
[0.1.0] — 2026-04-16¶
Added¶
- Core RAG pipeline:
async with RAG(...) as rag:context manager - Hybrid BM25+dense retrieval on every query, using Reciprocal Rank Fusion
- Three store backends:
InMemoryStore,LanceDBStore,PgVectorStore - LLM adapters: OpenAI, Anthropic, Ollama (local), and custom protocol
- Embedder adapters: OpenAI and local sentence-transformers
- Incremental indexing:
ingest()skips unchanged files via SHA-256 content hash ragwise initCLI — generates typed config templateragwise serveCLI — minimal HTTP API (Starlette)- Built-in eval:
rag.eval(),rag.eval_chunks(),assert_eval_passes()pytest helper - Production tracing via
LangfuseTracer - Advanced chunkers:
ContextualChunker(LLM-enhanced),HierarchicalChunker(child retrieve, parent generate) - LLM response caching (in-memory)
QueryConfig:top_k,include_citations,check_sufficiency- Cross-encoder reranker (opt-in)
- 170 tests — all passing