- Add --rag-debug flag to show retrieved chunk names and similarity scores
- Add explicit fallback notices when RAG indexing/query embedding fails
- Log RAG index/query metrics (duration, scores, top hit, token estimate)
- Normalize and cap chunk content for more stable prompt shape on small models
- Add hypothesis-continuity instruction for follow-up prompts
- Add retrieval scoring API and new tests for truncation/fallback/debug paths
- Add embed() to AIClient using Ollama nomic-embed-text via /v1/embeddings
- Add DEFAULT_EMBED_MODEL and embed_model field to AIConfig
- New rag_retriever.py: chunk_report(), EmbeddedChunk, retrieve() (pure-Python cosine)
- prompt_builder: add build_message_with_chunks() for RAG-aware follow-up prompts
- cli: add --no-rag flag, embed report chunks after collection, retrieve top-5 per question
- Graceful fallback to full-context if embedding model unavailable
- 16 new tests in test_rag_retriever.py (67 total, all passing)
- Add chromadb>=0.5 as optional [rag] dep in pyproject.toml
- README: add step 3 (pull nomic-embed-text), update Suggested Tooling table
