run-llama/rags

Leverages a two-agent architecture where a builder agent generates RAG pipeline configurations (system prompts, chunk size, embedding/LLM models) from natural language instructions, while the generated agent performs retrieval and agentic tool selection over indexed data. Supports flexible LLM backends (OpenAI, Anthropic, Replicate, HuggingFace) and embedding models, with configurable parameters like top-k retrieval, optional summarization, and chunk strategies—all editable through a Streamlit UI before querying. Built on LlamaIndex and enables rapid RAG prototyping without manual pipeline engineering.

6,534 stars. No commits in the last 6 months.