doc7
Turn documents into AI-ready Markdown with visual understanding
What is it?
What it is
A tool that converts various document formats (PDFs, Office files, images, etc.) into AI-ready Markdown using a local multimodal model, enabling searchable and structured content extraction without OCR or external services.
Why it exists
To provide an open-source, local alternative to cloud-based document processing services, eliminating per-page fees, OCR dependencies, and data egress concerns while maintaining full control over document processing infrastructure.
Who should use it
Developers, data scientists, AI researchers, and knowledge‑base builders who need to convert PDFs, Office files, images, and other visual documents into AI‑ready Markdown for RAG, semantic search, or document analysis, especially when they have access to a local or private multimodal model.
Who should avoid it
Users who rely solely on cloud document‑processing APIs, lack local VLM hardware, prefer a fully managed OCR service, or are uncomfortable with command‑line tools and model configuration.
How it works
A quick walkthrough in plain English
How doc7 works
Step 1 of 3
You interact with it
Open doc7, send a request, or connect it to your stack.
Features
Advantages
- Near-zero marginal cost when using local quantized models
- High accuracy in visual understanding benchmarks compared to standard OCR pipelines
- Maintains privacy by allowing entirely local, air-gapped inference
- Unified pipeline for diverse document types (text, tables, formulas, diagrams)
- AI-ready output optimized for RAG and semantic search
- Cross-platform support (macOS, Linux, Windows) via single binary or Docker
Disadvantages
- Requires external renderers (LibreOffice, MuPDF, Chromium) for certain file types
- Performance and quality are highly dependent on the chosen VLM
- High-resolution rendering and complex documents may require significant hardware resources
- Requires manual configuration for remote API endpoints
Installation
docker
docker compose pull && docker compose up --no-build
native
macOS or Linux: curl -fsSL https://raw.githubusercontent.com/magicrew/doc7/main/scripts/install.sh | bash Windows PowerShell: irm https://raw.githubusercontent.com/magicrew/doc7/main/scripts/install.ps1 | iex
compose
docker compose pull && docker compose up --no-build
FAQ
How do I install doc7 on macOS, Linux, or Windows?
Use the provided install scripts: macOS/Linux via `curl -fsSL https://raw.githubusercontent.com/magicrew/doc7/main/scripts/install.sh | bash`, Windows via PowerShell with `irm https://raw.githubusercontent.com/magicrew/doc7/main/scripts/install.ps1 | iex`.
Why should I use a local model instead of cloud document APIs?
Local models eliminate per-page/document fees, keep documents private, and leverage existing hardware (CPU/GPU) without recurring cloud costs.
What document formats does doc7 support?
doc7 supports PDF, Office files (DOCX, XLSX, PPTX), images (PNG, JPEG), EML/msg files, Jupyter notebooks, and more, processed through a unified visual-understanding pipeline.
How does doc7 perform compared to other tools like MarkItDown or Docling?
In benchmarks, doc7 achieved 15/15 verified facts on complex pages, outperforming MarkItDown (9/15) and Docling (3/15) on raster-only inputs.
How do I configure a remote model endpoint?
Run `doc7 setup config --base-url <remote-url> --model <model-id>` to specify a remote OpenAI-compatible endpoint and model ID.
What security measures are in place for document processing?
doc7 runs local renderers with user permissions, sanitizes inputs (e.g., HTML/email), and never stores API keys or document content in logs or artifacts.
Featured in Videos
YouTube tutorials and walkthroughs for doc7
Alternatives
Similar projects ranked by category, topics, and text overlap.
