v2.0.1-beta.3 — Read text in images and scanned PDFs (opt-in OCR), Windows & macOS

Read the text inside images and scanned PDFs (opt-in)

Slides full of screenshots and scanned PDFs used to come out nearly empty — the words were in pictures, not text. Turn on Read text in images and scanned PDFs in Settings and AI DocPrep now recovers that text on-device with a built-in OCR engine. Nothing is downloaded and nothing leaves your machine.

What it does when enabled:

• Slides, documents and spreadsheets: text found in embedded images is appended in a "Text found in images" section. Tiny icons are skipped and a logo repeated on every slide is reported once.
• Scanned PDFs: pages are read page by page. PDFs that already contain text are left exactly as before — OCR only kicks in when there's nothing to extract.
• Image files (PNG, JPEG, and friends): drop them directly to convert. Folder scans still skip images, so a photo folder never turns into a long OCR job.

It's off by default because it is much slower than normal conversion — roughly a second per image, so a picture-heavy deck can take a minute or two. Everything else runs at the usual speed whether OCR is on or off.

With OCR off, you can still see what you'd be missing: files that contain text inside images (or scanned pages with no text layer) get an "img unread" badge in the app and a note at the end of their Markdown, so you know which ones are worth re-running with OCR on. Dropping an image with OCR off tells you which setting to enable instead of quietly producing an empty file.

Notes

• The Windows installer is larger this release because the OCR engine and its models are bundled, so the feature works offline with no downloads.
• Conversion remains 100% offline.

Downloads

• Windows: AI DocPrep_2.0.1-beta.3_x64-setup.exe — signed.
• macOS (Apple Silicon): AI DocPrep_2.0.1-beta.3_aarch64.dmg — signed and notarized.

Still a beta — please open an issue if something converts badly or the app misbehaves.

v2.0.1-beta.2 — Signed Windows & macOS builds, new formats, opt-in ZIP

Both platforms are now code-signed. On Windows, no more "Windows protected your PC — unknown publisher"; on macOS, the app is signed with a Developer ID and notarized. Installer, app, and conversion engine are all signed and timestamped.

Note: Windows SmartScreen reputation builds up with download volume, so the first few installs may still show a milder warning before it clears entirely.

New formats

• EPUB — turn ebooks into Markdown notes
• Jupyter notebooks (.ipynb) — markdown cells, code, and output
• Outlook messages (.msg) and legacy Excel (.xls)

ZIP archives, opt-in

Turn on Convert ZIP archives in Settings to convert the documents inside an archive. It's off by default, and deliberately conservative when on: contents are filtered to supported formats, nested archives are skipped, and size limits apply. Anything skipped is listed at the end of the output so nothing disappears silently. Archives convert when you drop them as files — folder scans don't open them.

Leaner install

Dependencies were narrowed to only the formats this app actually converts, removing optional packages that were never used. Conversion remains 100% offline.

Downloads

• Windows: AI DocPrep_2.0.1-beta.2_x64-setup.exe
• macOS (Apple Silicon): AI DocPrep_2.0.1-beta.2_aarch64.dmg

This is still a beta — please open an issue if something converts badly or the app misbehaves.

v2.0.1-beta.1 — Signed macOS Beta

AI DocPrep v2.0.1 Beta 1 — Native shell, faster engine, on-device PII redaction, and official Apple Developer ID code-signing.

Note: This is a limited-time beta release. Future signed releases will be available directly on the Mac App Store.

Highlights in v2.0.1 Beta 1

• Official Developer ID Signing: Built and signed with an official Apple Developer ID certificate and Hardened Runtime, eliminating Gatekeeper "damaged app" warnings on macOS.
• Drop-Queue Workflow: Drop any mix of files and folders into the window; watch conversions happen live with token savings estimates (e.g. −81% vs raw).
• 3 Local Redaction Engines: Instant regex patterns, on-device spaCy NER (en_core_web_sm for names/orgs/locations), or local Ollama LLM models. Everything stays 100% offline.
• Tauri 2 & Svelte 5 Native Shell: Ultra-lightweight UI with low memory usage and fast launch times.

Installation (macOS)

1. Download AI DocPrep_2.0.1-beta.1_aarch64.dmg below.
2. Open the .dmg and drag AI DocPrep into your /Applications folder.
3. Open AI DocPrep directly from your Applications folder or Spotlight!

Verification & Integrity

• Architecture: macOS Apple Silicon (aarch64)

v2.0.0-alpha.1 — Native rebuild (Windows & macOS pre-alpha)

A ground-up rebuild of AI DocPrep — native shell, faster engine, redesigned interface. This is an early pre-alpha; I'd love your feedback.

Drop files or whole folders, get clean token-efficient Markdown for your LLM or PKM vault. Everything runs 100% offline.

Installing (Windows)

1. Download AI DocPrep_2.0.0-alpha.1_x64-setup.exe below and run it.
2. Windows SmartScreen will warn you — "Windows protected your PC / unknown publisher." This is expected: the app isn't code-signed yet (that's coming). Click More info → Run anyway to proceed. If you'd rather not, wait for a signed build.
3. Launch AI DocPrep from the Start menu.

Installing (macOS)

1. Download AI DocPrep_2.0.0-alpha.1_aarch64.dmg below.
2. Open the DMG and drag AI DocPrep into your /Applications folder.
3. macOS Gatekeeper will block launch ("damaged" / "unidentified developer") on the unsigned binary. Bypassing this:
   • Open System Settings -> Privacy & Security -> Security.
   • Click Open Anyway next to the block warning.
   • Alternatively, right-click the application, select Open, and then click Open in the confirmation dialog.

What's new since 1.x

• Drop-queue workflow — drop any mix of files and folders, watch each one convert live, with folders expanding into their individual files.
• Token savings, shown honestly — every conversion reports how much smaller the Markdown is than feeding the raw file to an LLM (e.g. −94% vs raw).
• On-device redaction — strip PII before anything leaves your machine: instant pattern matching, an offline AI model for names/orgs/locations, or your own local Ollama.
• Much faster startup and a redesigned paper-and-ink interface.

Known limitations

• Not code-signed yet, so OS security warning screens appear on first launch.
• Scanned/image-only PDFs have no text layer to extract (no OCR yet).

Feedback

This is pre-alpha — rough edges are expected and reports are genuinely useful. Please open an issue with what you converted, what you expected, and what happened. Crashes, weird output, and confusing UI moments are all worth flagging.

v1.2.0 — Redesigned app, token savings, headless CLI

A ground-up redesign of the app around one simple flow: drop → convert → results.

Redesigned app

• Drop anything, anywhere — files, folders, or a mix, dropped anywhere on the window (or added via buttons/⌘O). No more File/Folder mode switch, and multi-file drops now work.
• One Output menu replaces the combine toggles: Individual .md files, Individual + combined file, or Combined file only.
• Results you can act on — per-file status with token counts, then Show in Explorer/Finder, Copy Markdown, or start a new batch. The old activity log lives under File → Activity Log.
• Settings that stick — every option (output, YAML, redaction engine, custom prompt/terms) is saved and restored between launches.

Token counts and savings

• Every converted file shows an estimated token count — what it will cost in an LLM's context window.
• Where an honest baseline exists, you also see savings vs the raw file: Office formats are compared against their internal XML, HTML/VTT against their raw text, and PDFs against the ~2,000 tokens/page providers charge for raw PDF uploads.

New: headless CLI

docprep_core.py runs the whole engine without the GUI — great for scripts and automation:

python docprep_core.py ./research --output combined-only --redact
python docprep_core.py ./docs --json   # one JSON event per line

Fixes and safer behavior

• File conflicts now default to keep both (file (1).md) instead of silently overwriting.
• The folder combiner only merges files it just converted — your pre-existing Markdown notes are never swept into the combined output.
• Office temp/lock files (~$...) are skipped automatically.
• Fixed intermittent crashes from cross-thread UI updates.
• Local Ollama redaction requests are serialized so they no longer time out under parallel conversion.

Downloads

• Windows: AIDocPrep_Setup.exe below.
• macOS: coming shortly to this release.

AI DocPrep v1.0 Launch

Preview with privacy local model. UI Refresh.

AI DocPrep v1.1 Preview

Preview new 1.1 privacy features