Introducing Wonder Reader: a free Mac reader with VoiceOver-focused navigation and local AI

Hi everyone,

I became frustrated with the cost of Voice Dream on the Mac and with an update experience that, for me, did not justify an ongoing subscription. Its US Mac App Store listing includes a $49.99 annual subscription, alongside other subscription prices. I wanted a reading app I could use without another recurring bill, so I decided to build one and share it with the community for free.

I am the developer of Wonder Reader, a native Mac reading app designed around keyboard access and VoiceOver. Reading, text-to-speech, OCR, translation and summaries are all free in this direct-download release. There is no account requirement or subscription. Wonder is named after my guide dog.

The current release is 0.1.8 (41), a software-only download of about 47.2 MB.

Download and installation instructions:
https://dev.mk-musician.com/wonder-reader/

You need an Apple Silicon Mac (M1 or later) running macOS 14 or later. Intel Macs are not supported. Download the ZIP, unzip it, move Wonder Reader.app into Applications and open it. The release is Developer ID signed and notarized by Apple. No UDID registration is needed for this Mac app.

What you can read

You can import DRM-free EPUB books, PDFs, DOCX documents, Markdown, plain text, HTML and RTF. EPUB books retain the table of contents supplied by the book. Reading positions are saved so you can return to where you stopped, and bookmarks provide another way back to useful passages. Original documents remain separate from generated translations and summaries.

For PDFs with usable text, the app extracts that text. Scanned pages can be processed with Apple's local OCR. OCR accuracy depends on the scan and layout; check important names, numbers and tables against the original. Structured table recognition uses newer system capabilities on macOS 26; older supported systems do not offer the same table experience.

Listening and accessibility

Wonder Reader uses the system voices available to Mac apps, with controls for voice, speed, pitch and volume. You can follow VoiceOver's speech settings or choose independent reading settings, including preferences for your main language and other languages.

The reading area uses native text navigation. You can pause, move the text insertion point and resume from the new position. I am particularly interested in feedback on how this behaves with different VoiceOver cursor-tracking settings and braille setups. I do not want to assume that one configuration proves every configuration works well.

Some useful shortcuts while using the app:

- Command-O: import a document.
- Command-1: library; Command-2: table of contents.
- Command-3: original text; Command-4: processing results; Command-5: bookmarks.
- K: play or pause in the reading area.
- J / L: previous or next sentence; Shift-J / Shift-L: previous or next paragraph.
- Command-J / Command-L: previous or next chapter.
- Command-backtick: AI assistance.
- Z / X / C: OCR, translate or summarize the current chapter or page; add Shift to process the whole document.
- Slash: hear processing progress.
- Command-Shift-X: stop processing.
- Command-comma: Settings.

The welcome guide is also available from Help. The interface supports English, Simplified and Traditional Chinese, Japanese, Korean, French, German, Spanish and Portuguese.

Optional AI that runs on your Mac

Ordinary reading, system speech and Apple OCR work without downloading an AI model. The app download does not include model weights. Translation, summaries and model-assisted interpretation require a separate model download in Settings. You do not need Python, Homebrew or a separately configured AI server.

There are four optional model tiers:

- Low: Qwen3 1.7B Q3_K_S, about 867 MB.
- Mid: Qwen3 1.7B Q4_K_M, about 1.11 GB.
- High: Qwen3 4B Instruct 2507 Q4_K_M, about 2.50 GB.
- Advanced: mlx-community/Qwen3.5-4B-4bit, about 3.06 GB.

The app displays download sizes and lets you download, select or remove models. Model downloads need an internet connection; AI processing runs locally after the required model is installed. Download size and running memory requirements are different things.

You can choose output languages and summary detail, then listen to the result without replacing the original text. These are small local models, and translations or summaries can be awkward or incorrect. Larger models are not a guarantee of accuracy, especially for literary writing, terminology, numbers and long documents.

What is new in this release

The app now distinguishes model startup and input preparation from generation. Slash reports the current stage, percentage and elapsed time; startup and processing errors are announced and can be queried again. Model downloads display speed and continue when Settings closes. Quitting the app stops downloads, but starting again can resume retained data. Advanced now also retains partial downloads, and incomplete downloads can be removed from model management.

I would love feedback

This is an early release. An intermittent Command-Escape problem in the results area remains under investigation. VoiceOver navigation in Settings and Trash and extended use still need more real-world feedback. This release had real inference checks with High, not a quality benchmark of all four models.

I especially welcome reports about keyboard navigation, VoiceOver focus, speech interruptions, model downloads, translation quality and documents that do not import cleanly. Please tell me your macOS version, app version, what you tried and what happened. There is no need to share a private book or document to report a problem.

My aim is to make an accessible reading tool that people can actually use without a subscription barrier. If you try it, please let me know what works and what needs attention.

Disclosure: I used AI assistance to help draft this English introduction from my own requirements and experience.