Documind is an open-source document processing tool that leverages AI vision models to extract structured data from unstructured documents. Maintained at github.com/DocumindHQ/documind, it bridges the gap between raw document content and the clean, typed JSON that applications actually need.

The key innovation in Documind is schema-driven extraction. Instead of getting back a blob of text from a document, you define exactly what fields you want — account numbers, transaction dates, line items, totals — with their expected data types and descriptions. Documind then uses AI vision models to locate and extract those specific fields, returning structured JSON that matches your schema. This makes it practical for automating document workflows where you need specific data points, not just raw text.

Under the hood, Documind converts documents to images using Ghostscript and GraphicsMagick, then sends them to an AI vision model for analysis. It supports OpenAI’s GPT-4 Vision as well as local models through Llama 3.2 Vision integration, giving teams the option to keep sensitive documents on-premise. The tool also supports auto-generated schemas — point it at a document and it will infer what fields are available.

Beyond JSON extraction, Documind can convert documents to Markdown format, which is useful for feeding document content into LLM pipelines. It includes template schemas for common document types like invoices, bank statements, and receipts, so you can start extracting data without writing custom schemas.

A skill wrapping Documind gives an AI agent the ability to process incoming documents and extract exactly the data it needs. The agent could parse invoices for accounting workflows, extract form data for processing, or convert documents into structured formats for database ingestion. Install via npm install documind. The project requires Node.js 18+, Ghostscript, and GraphicsMagick as system dependencies. Documind is MIT-licensed with active development.