https://github.com/ggozad/haiku.rag is a deceptively simple and competent RAG library. It scales for endusers very well. It is self contained. And works in < 1 minute without any additional sit ups.

Some features:

• utilizes docling to index with reasonable chunker strategy
• performs rag: search, ask
• has multi-agent agentic RAG oriented research feature
• runs from command line or easily embedded into python application
• tight integration with pydantic_ai
• using lancedb it is serverless and scales using s3 buckets as data sources

Pretty clever.