vlmrun-hub

Why we included this project

Most teams building document or image extraction end up writing their own Pydantic schemas and debugging them against real-world inputs. VLM Run Hub removes that step by shipping a catalog of pre-built models for common domains: invoices, receipts, driver's licenses, passports, bank statements, medical insurance cards, and even sports game states. Point a vision-language model at an image and you get validated, strongly-typed output rather than free-form text. The schemas are model-agnostic, so the same definition works with OpenAI, Gemini, or other structured-output providers, and they plug into the VLM Run SDK or Instructor-style clients. If you do document processing, accounting automation, or other visual ETL, the ready-made domains are a practical starting point, and the catalog format makes it easy to contribute new schema types as needs grow.