magicrew/doc7

What it solves

doc7 is a document conversion tool that transforms complex documents—including PDFs, Office files, scans, screenshots, charts, and diagrams—into AI-ready Markdown. It solves the problem of losing structural and visual information (like formulas, tables, and diagram relationships) during traditional OCR or text extraction, making documents more searchable and reason-able for AI models.

How it works

Instead of relying on a traditional OCR stack or file-specific parsers, doc7 uses a full-page visual understanding approach. It renders document pages as images and processes them through an OpenAI-compatible multimodal (vision) model. This allows the tool to reconstruct the page content, including visual elements, as Markdown. Users can run these models locally (via LM Studio or Ollama) to maintain privacy and eliminate per-page API costs.

Who it’s for

• AI Developers: Those building RAG pipelines who need high-quality, structured Markdown from diverse document formats.
• Privacy-Conscious Users: Individuals or organizations that want to process documents using local vision models to avoid sending data to cloud APIs.
• Self-Hosters: Users looking to eliminate recurring document-parsing fees from cloud providers.

Highlights

• Multimodal Pipeline: Uses a single visual-understanding pipeline for all formats (PDF, Office, images) instead of separate OCR/layout models.
• Local Model Support: Compatible with any OpenAI-compatible multimodal model, including local deployments via Ollama or LM Studio.
• Zero Marginal Cost: No per-page fees or document credits when using local hardware.
• Robust Recovery: Supports resuming failed page conversions and reprocessing specific pages with different models.
• Versatile Interface: Provided as a cross-platform CLI, Go SDK, MCP tool, and an asynchronous HTTP service.