Architecture

Knowledge Ingestion

Knowledge Ingestion prepares source material for retrieval. It loads files and folders, chunks text, attaches metadata, creates embeddings, and stores them in a vector store.

Retrieval Flow

Retrieval Flow turns a user question into an embedding and finds the most relevant stored chunks. Those matches provide the bounded context used to ground an answer.

Answer Orchestration

Answer Orchestration coordinates retrieval, context construction, and the configured chat provider through the RAG API. It returns an answer together with retrieved sources and the exact context excerpts supplied to the chat provider. SQLite and PostgreSQL support opt-in keyword/vector rank fusion; the cosine score and fusion score remain separate.

Future Applications

Future Applications are the product capabilities built on top of PyAIStack’s RAG foundation. Chat apps, search, assistants, automation, and workflows are planned directions rather than current modules.