v0.3.0 — Phase 3: Advanced Configuration & Deployment

All Phase 3 changes since Phase 2 (v2.0.0), plus package version sync.

Configuration

• Embedding and LLM — Embedding model and LLM provider configurable via env or config (e.g. swap OpenAI for other providers).

MCP server

• Resources — List of indexed documents as read-only resource (oracle-rag://documents).
• Prompts — Pre-built prompt with parameters: “Ask about this document” (ask_about_documents).

Retriever abstraction

• LangChain Retriever — RAG pipeline uses store.as_retriever() instead of direct query_index.
• run_rag() — Accepts optional BaseRetriever (e.g. for reranking) directly.

Error handling

• Corrupted PDFs — load_pdf_as_documents catches PyPdfError/OSError and re-raises a clear ValueError.
• Zero retrieval — No chunks → clear message without calling LLM.
• LLM failure — Rate limit, timeout, or other errors → short message instead of traceback.

Testing

• Multi-PDF integration — Index two PDFs with different tags; query by document_id and tag; assert cross-document retrieval and correct sources.

Package

• Version — pyproject.toml set to 3.0.1.