OpenAI RAG Starter Kit with File Search and Chat UI (github.com)

OpenAI’s new config-first RAG starter kit provides a one-command-to-run path for building a grounded knowledge-retrieval app by combining OpenAI File Search, ChatKit UI, and OpenAI Evals with a pluggable ingestion/retrieval/evaluation toolkit. The repo ships CLI scaffolds, YAML-first configs, typed pipelines (Pydantic + Typer + FastAPI), and a Vite/React + Tailwind frontend using @openai/chatkit-react, so you can ingest documents, run a chat UI that returns citation-linked answers, and run synthetic or user-defined evals end-to-end (quickstart shown with make targets and rag CLI commands). Requirements: Python 3.10+, Node 18.18+, and an OpenAI API key with File Search/Evals access. Technically, the kit is built for experimentation and production iteration: multiple vector store backends (managed OpenAI File Search or custom adapters—Qdrant example included, plus plugin hooks), configurable chunking strategies (recursive, heading, hybrid, xml_aware, or custom), and four optional query stages (LLM expansion, HyDE synthetic docs, similarity filtering, and LLM reranking). Synthesis is configurable (model choice, system prompts, structured outputs, reasoning effort), and the eval harness can synthesize datasets from ingested chunks or run user datasets, with local grading or mirrored OpenAI Evals dashboards. The result is a reproducible, modular RAG baseline that accelerates building, tuning, and validating grounded QA/retrieval systems.