Introduction

rag-server is a Go-based backend for Retrieval-Augmented Generation (RAG). It provides APIs to ingest documents, perform hybrid retrieval (vector + full-text), and optionally call a chat-completion model to answer questions.

Key capabilities from the current codebase:

• RAG retrieval: Vector similarity via pgvector plus full-text ranking via PostgreSQL tsvector and zhparser.
• Ingestion pipeline: Sync markdown content from Git repositories, chunk it, embed it, and upsert into Postgres.
• Model-agnostic: Uses OpenAI-compatible HTTP APIs for embeddings and chat completions (also supports Ollama/Chutes providers).
• Operationally simple: Single stateless service with Postgres as the only required backend.
• Auth middleware: Optional JWT validation with cached token lookups in Postgres.

This repository ships two main binaries:

• rag-server: The HTTP API service.
• rag-cli: A CLI for syncing and ingesting markdown content into the server.

Related tools (optional): ingest (batch ingestion) and ragbench (benchmark runner).