Skip to content

Folge Vision Publishing Pipeline

Version: 2026.8.2 | License: Apache 2.0 | Author: Michael Hunsaker

An automated documentation publishing pipeline that enriches Folge guide exports with Vision AI-generated accessibility metadata, then publishes to every format the installed pandoc supports (70+ writers) including PDF/UA-compliant PDFs, DOCX, HTML, EPUB, LaTeX, Typst, and more.

What It Does

  • Export

    Take your guide and screenshots from Folge

    Learn more

  • Enrich

    Vision AI generates alt text, descriptions, OCR, and UI control detection

    Learn more

  • Merge

    Combine authored content with vision data deterministically

    Learn more

  • Validate

    Ensure schema compliance and content quality

    Learn more

  • Render

    Generate Markdown with embedded accessibility metadata

    Learn more

  • Metadata

    Embed title, author, subject, keywords, language, and security settings into every format

    Learn more

  • Publish

    Convert to every format the installed pandoc supports, with PDF/UA compliance

    Learn more

  • GUI

    An accessible, browser-based NiceGUI front end for the whole pipeline

    Learn more

Key Features

  • Eight AI providers -- ollama (default), lmstudio, jan, llamacpp, openrouter, openai, gemini, anthropic
  • Project folders -- each guide lives in ~/Documents/FolgeProjects/<project>/ (guide JSON + images/ + output/), so you can keep many guides organized
  • Accessible GUI -- a WCAG 2.2 AA NiceGUI front end (uv run folge-gui) for setup, individual steps, and the full pipeline
  • 70+ output formats -- every writer the installed pandoc supports (PDF, DOCX, HTML, PPTX, GitHub Markdown, Typst, AsciiDoc, Beamer, CommonMark, GFM, MultiMarkdown, DocBook, EPUB, ODT, RST, LaTeX, and more)
  • Accessibility-first -- WCAG 2.1 AA, ARIA, PDF/UA, DOCX accessibility support
  • Accessible document metadata -- auto-generated title, author, subject, keywords, language, tags, bookmarks, and copy-permissive security embedded into all formats
  • PDF page orientation -- Letter portrait (default) or Letter landscape
  • Self-contained output -- HTML and EPUB embed all resources
  • Custom fonts -- Atkinson Hyperlegible Next (text) and AtkynsonMonoNerdFont (code/monospace)
  • Deterministic -- Same input always produces same output
  • Separation of concerns -- Authored content stays separate from AI enrichment
  • Pre-built binaries -- Coming Soon (single-file executables for Windows, macOS, and Linux)

Quick Start

git clone https://github.com/mrhunsaker/Folge_Accessibility.git
cd Folge_Accessibility
uv sync

# Pull the vision model (if using ollama)
ollama pull qwen2.5vl-8k:latest

# Run the full pipeline (every supported format)
folge-cli pipeline --project my-first-guide

# Or publish specific formats
folge-cli publish --project my-first-guide pdf,docx,html,epub,typst

Not yet available

Pre-built binaries are not yet released. Use the source installation in the meantime. Check GitHub Releases for updates.

Full getting started guide