Architecture Overview

High-level flow

Client
  -> rag-server (Gin HTTP API)
     -> Postgres (pgvector + full-text)
     -> Embedding Provider (OpenAI-compatible)
     -> Chat Completion Provider (OpenAI-compatible)
     -> Git repos (via go-git, optional)

Data flow (RAG query)

1. Client POSTs a question to /api/rag/query.
2. Server embeds the question using the configured embedder.
3. Server queries Postgres for vector + full-text candidates.
4. Scores are blended and optionally reranked.
5. Top documents are returned as chunks.

Data flow (Ingestion)

1. rag-cli (or /api/rag/upsert) embeds content.
2. The server ensures the schema exists.
3. Chunks are upserted into Postgres.