Chandra is Datalab’s open-source OCR model for the documents that break ordinary OCR: complex tables, multi-column forms and handwriting, all returned with their layout intact as clean structured output. From the team behind Surya and Marker, it has gathered about 12,500 stars on GitHub since its 2026 release.
Traditional OCR engines output a stream of characters and leave you to reconstruct the document. Chandra is a vision-language model trained to read a page the way a person would, preserving reading order, table structure, form fields and handwritten entries, and emitting Markdown, HTML or JSON that downstream systems can actually use. That makes it suited to the messy end of document processing โ scanned contracts, invoices with nested tables, historical forms, mixed print and handwriting โ where template-based extraction fails. The model is released openly and can be run locally through standard inference stacks, while Datalab also operates a hosted API for teams that prefer not to run models themselves. Chandra continues the lab’s line of document-AI tools that includes the Surya OCR toolkit and the Marker PDF converter, both widely used in RAG ingestion pipelines.
Chandra’s model and code are free and open source under the Apache-2.0 license for self-hosting. Datalab also sells hosted API access with paid plans, so the overall offering is freemium: free on your own infrastructure, paid if you use their cloud.
Chandra is for teams extracting structured data from difficult documents โ financial forms, scanned archives, handwritten paperwork โ and for RAG builders who need faithful document conversion that keeps tables and layout instead of flattening them into garbled text.