RAG track
This track walks from RAG concepts in ArcFlow through document ingest, agent wiring, and a first look at hybrid retrieval. You use VectorStore.ingest, attach MemoryConfig to agents, and verify retrieval through workflow traces.
What you will learn
| Lesson | Topic |
|---|---|
| 01 RAG concepts in ArcFlow | Ingest, embed, retrieve, inject; how namespace ties the pipeline together |
| 02 Ingest documents | VectorStore.ingest(namespace, key, text) and chunk counts |
| 03 Retrieval and agent wiring | Query runs, MemoryRetrieved events, end-to-end Q&A |
| 04 Hybrid retrieval intro | Dense plus sparse weights, ARCFLOW_QDRANT_HYBRID, rerank basics |
Prerequisites
Complete the Memory track, especially 03 Vector memory setup. You need the Python SDK built (Install and build) and Qdrant reachable at ARCFLOW_QDRANT_URL for the full ingest-and-query path. Stub embedding (stub/8 or stub/384) is enough for local dev without a paid embedding API.
Primary example scripts:
| Branch | Script |
|---|---|
| Current branch | RAG chatbot walkthrough |
| Examples restructure branch | examples/rag/memory_guide_qa.py (ingest plus query in one file) |
How these lessons are structured
Every page follows the same sections: Before you start, Concept, Example, Verify, and Next.
After this track
| Goal | Next document |
|---|---|
| Tutorial with verification checklist | Track C: RAG and vector memory |
| RAG chatbot walkthrough | RAG chatbot example |
| Hybrid tuning deep dive | Hybrid retrieval and reranking |