// the find
run-llama/rags
Build ChatGPT over your data, all with natural language
RAGs is a Streamlit app that lets you configure a RAG pipeline through natural language conversation — you describe your data source and retrieval preferences, and a builder agent turns that into a working LlamaIndex RAG agent. It's aimed at people who want to prototype 'ChatGPT over my docs' without writing glue code. Think of it as a UI wrapper around LlamaIndex's agent configuration, not a production system.
The builder-agent approach is genuinely clever — using an LLM to configure another LLM pipeline is a neat abstraction that lowers the barrier significantly. The config view that exposes and lets you override generated parameters (top-k, chunk size, system prompt) is the right UX call; you're not locked into what the builder decided. Supporting multiple LLM backends (OpenAI, Anthropic, Replicate, local HuggingFace) at the generated-agent layer means you're not OpenAI-only at runtime. The separation between a 'builder agent' and the 'generated RAG agent' is a clean architectural decision for a project of this scope.
Last commit April 2024, and this space moves fast — LlamaIndex has had breaking API changes since then, so expect pain on install. The README itself warns about cache breaking changes between versions, which is a red flag for reliability. Supports only a single local file or a single web page as data sources — no folder ingestion, no database connectors, nothing a real use case would actually need. The README quietly admits GPT-4 is required for reliable builder behavior ('we couldn't get Claude to work'), which means you're paying for the expensive model just to configure your pipeline, not to answer questions.