Fido
Purpose
Making publications accessible is skilled, time-consuming work. Fido is a tool from the DAISY Labs that uses AI to take on much of the heavy lifting — converting PDFs, describing images, detecting languages, analyzing structure and generating metadata.
Background
Fido launched on the DAISY website in late April 2026, the result of twelve months of experimentation with AI technologies and workflows. Its functionality, design, and outputs were shaped throughout by members of the DAISY Consortium AI Special Interest Group. Fido has been valuable as we develop ideas for how AI can be used in the production, remediation and conversion of accessible publications. The app is an incubator for ideas and evaluations, hence the phrase ‘from the DAISY Labs’. This is not a production standard app with long-term support like other software tools from DAISY.
The app has also been increasingly used by participants in our Training and Technical Assistance projects, especially for PDF conversion and image descriptions. Partner organizations have been using Fido to help their production of accessible books in English, French, Spanish, Hindi, Arabic and more, and they have reported impressive time savings. Books into the hands of people with print disabilities, faster!
The feedback we are hearing from these partner organizations:
- Fido isn’t a silver bullet – but is a serious shortcut.
- Tasks that took days can now take hours.
- Fido makes their people faster, but not redundant.
Who is Fido for?
We’ve made the experimental app available from the DAISY website because if you make, remediate or convert accessible publications you might also find Fido useful. If you use a workflow based on Microsoft Word for Windows, then you’ll find the Word add-in especially helpful.
To use most of Fido’s cloud AI features you will need either an unlock code from DAISY or at least one API key from a cloud service. Fido also includes built-in models that run on your computer and need neither.
Features
Choose your AI. A step by step setup wizard helps you pick and configure the AI services and models that suit your content, budget and privacy needs. Fido works with cloud services from many providers, including OpenAI, Anthropic, Google, Mistral, xAI, Moonshot, Alibaba and Microsoft Foundry (use models deployed in your organization’s own Azure). Aggregators such as OpenRouter, Requesty and Together AI give access to many more. Fido can also run built-in models on your own computer, so some features work without sending your content to a cloud service.
Convert from PDF to other formats (Word, EPUB, HTML). Works with both image only and searchable PDFs. Extract text, headings, lists, images, tables and math content. Many conversion services are supported, including Marker, Mistral, MuPDF, Gemini, Claude and ChatGPT models. PaddleOCR, GLM-OCR and Granite-Docling run on your computer.
Describe images in Word, PowerPoint, PDF, EPUB and InDesign documents. Use cloud AI services such as ChatGPT, Claude, Gemini, DeepSeek, Mistral and more, or a built-in model that runs on your computer. Adjust the prompts to meet your content and editorial needs. Use custom prompts for different image types (Fido recognizes more than 100 image classifications). Generate extended descriptions (Word only).
Mark up phrases and paragraphs in different languages in Word or EPUB documents by using AI to detect words in different languages, ignoring proper nouns and ‘loaned’ words that should remain unchanged.
Analyze heading structure to detect missing or misclassified heading levels. This feature is intended to help improve conversions from scanned documents or converted PDFs and works in Microsoft Word.
Generate metadata from your Word document by using AI to read the document to extract or generate fields such as title, authors, publisher, ISBN, language, keywords and summary.
Responsible use
AI services make mistakes. Always review and verify the output.
Each AI service has its own terms and conditions governing how data is processed and stored. You are responsible for ensuring your use of any AI service complies with its terms. The DAISY Consortium is not responsible for the services integrated into this app.
Can you help?
We welcome bug reports and feature suggestions which can be submitted using the GitHub issues button in the About Fido dialog.
Downloads
Latest version
Fido 0.9.9 was released 2 November 2026.
Added
- Step by step wizard to choose and configure AI.
- Built-in PDF conversions that run on your computer (PaddleOCR VL 1.6, GLM-OCR, Granite-Docling).
- Built-in AI models (Ministral 3 8B and Qwen3-VL 8B) that run on your computer for image description and text features (metadata, headings, language).
- Microsoft Foundry (Azure AI Foundry): use GPT, Claude and other models deployed in your organisation’s own Azure.
- GPT-6.1 Sol and GPT-6 Luna via OpenAI, OpenRouter and Microsoft Foundry.
- Claude Sonnet 5.5 via Anthropic, OpenRouter and Microsoft Foundry.
- xAI (Grok 4.7, Grok 4.3) and Moonshot (Kimi K3, Kimi K2.6).
- Qwen3.8 Flash and Qwen3.6 Plus via Alibaba.
- WaveSpeed for text-to-speech (ByteDance Seed Speech TTS 2.0).
- PDF conversion via more multimodal models (GPT-6.1 Sol, GPT-5.5, GPT-5.4, GPT-5.4 Mini, Sonnet 5.5).
- PDF conversion via aggregators OpenRouter, Requesty and Together AI.
- Keep alt text already in the PDF.
- Extract embedded images from the PDF, with an alternative to use AI vision.
- Join split headings in PDF conversion and the Headings tool.
- Built-in prompt updates apply automatically if you have not edited them.
- Offer to update edited prompts when built-in prompts improve.
Changed
- API keys are now saved in the system’s credential store (Credential Manager on Windows, Keychain on Mac).
- PDF conversion runs many pages at once, slowing down if rate limited.
- PaddleOCR PDF conversion now runs on your computer.
- Rebuilt analysis algorithm and prompts in Headings tool.
- Heading styles are learned from sample pages across the whole document.
- PDF pages blocked by a content filter are recovered using various strategies.
- Simpler PDF conversion progress dialog.
- Improved estimated time remaining for PDF conversion.
- Improved screen reader progress announcements.
- Model descriptions are clearer and more consistent, and prices come from LiteLLM’s price list where available.
- Settings fits on smaller screens, with tabs that scroll when needed.
Fixed
- Cancelling Claude PDF conversion could hang.
- Word output failed when PDF alt text contained hidden characters.
- Figure images could include the caption or most of the page.
- Ugly white space in PDF settings dialog when a conversion service hides its options.
- Fasttext language detection on macOS.
- Language detection error on macOS.
- Numbered lists from some PDF conversions showed a bullet as well as the number in Word.
- Truncated French and Spanish Setup tab name.
- Updates failed without a message when another program was using Fido’s files.
- Print page marker prefix and suffix no longer gain extra leading and trailing spaces.
Removed
- Tensorlake PDF conversion: Tensorlake no longer offers document parsing.
- PaddleOCR PDF conversion via cloud services: reliability issues.
Older versions
Fido version 0.9.8
Fido 0.9.8 was released 2 October 2026.
Added
- PDF conversion supports mixed page numbering styles in one publication (eg roman, decimal, unnumbered).
- PDF conversion can use existing page labels if the source has them.
- PDF conversion can detect page numbers using computer vision.
- User can set document language for PDF conversion.
- Language of PDF can be detected automatically.
- Page number detection for Khmer, Thai, Lao and Myanmar digits.
- Add print page-break markers to an EPUB from a corresponding PDF.
- Increased parallel workers for faster PDF conversion, automatically scaling back if the API rate-limits requests.
- Increased parallel workers for image description, automatically scaling back if the API rate-limits requests.
- Claude Opus 5.5 model availability.
Changed
- Gemini PDF conversion improved image extraction.
- Gemini PDF conversion retries failed chunks.
- Gemini PDF conversion retries chunks rejected due to RECITATION.
- Command line arguments for PDF conversion have been updated to support new features.
Fixed
- Install Word addin from Fido setup does not work.
- Orphaned keyboard focus when launching language and headings utilities from Word addin.
- Generate metadata for empty fields does not work.
- Language markup cannot be applied to some paragraphs in Word.
Fido version 0.9.7
Fido 0.9.7 was released 1 September 2026.
Added
- Support for dark mode
- Language detection for HTML files
- AI image check for image and alt-text quality
- Image tool QA pane, verdict badge, and silent AI sniff while browsing or when the tool opens
- Support for H6 page number markup
- Page number suffix for PDF and EPUB conversions
- Page number styles for Arabic-Indic, Urdu-Indic, Devanagari, and Bengali digits
- Integrated CheckMate file checker into Fido main menu
- Support for Google Gemini Flash 3.7
- Command line support for Image report with optional AI analysis
- Convert tool now accepts docx, epub, html, odt and md files for simple conversion with Pandoc
- Metadata from EPUB is preserved when converting to Word
- Page markup is preserved when converting EPUB to Word
- Document language is preserved when converting EPUB to Word
- Users can edit localisation strings and add new languages with AI translation
Changed
- Converted Word files use the current Word format rather than Word 2007
Fixed
- Images with unusual colorspace are not extracted correctly from PDF
- PDF image extraction better supports vector objects
- Batch describe all images does not retain alt for images marked as decorative
- Cropped, stretched and rotated images are not extracted correctly from Word
Fido version 0.9.6
Fido 0.9.6 was released 4 August 2026.
- Added – Google Gemini Flash 3.6 for PDF conversion, image description and text processing.
- Added – Claude Opus 5 for PDF conversion, image description and text processing.
- Added – OpenRouter MoonshotAI Kimi K3 for image description and text processing.
- Added – image classifications and prompts from Benetech
- Added – MAI-Voice-2-Flash for TTS
- Added – Qwen-Audio-3.0 TTS with full voices and steadier long-form synthesis
- Added – page number style (Arabic or Roman)
- Fixed – XML repair for complex scripts in Word documents
- Fixed – ElevenLabs TTS stopped working due to deprecated models