RAOpen source
RAGFlow
Connect document parsing, chunking, and retrieval to build answers with source references.
Multimodal Retrieval & RAG
ColPali
Encode document page images as multi-vectors and retrieve visual pages with text queries.
At a glance
- Input
- Page images and text queries
- Output
- Embeddings and ranked results
- Provenance
- Page mapping must be maintained by the application
- Scanned pages
- Page images can be indexed
- Deployment
- Self-hosted
- Availability
- Open source
- Costs to plan for
- Compute and storage costs
Limitations to check
This is a retrieval component. You still need an index, page identifiers, and a model or application that produces answers.
Build a meaningful sample
Include your hardest pages: low-quality scans, missing fields, dense tables, and evidence split across pages. Validate output correctness separately from citation accuracy.