Skip to content
PocketWebToolsPocketWebTools
Get the Mac app
PocketWebTools for Mac

Changelog

Every release and model refresh, dated. Every entry here was, or will be, a free update for everyone who bought the app.

  1. Version 1.0: the app is on sale

    • PocketWebTools for Mac 1.0.0 is out. Everything in the entries below is in it: chat with open models up to 35B parameters, transcription with speaker labels, photo upscaling, OCR to markdown and searchable PDFs, background removal, Clip, Edit, and the everyday utilities from the website.
    • One payment, no subscription. A license key works on 3 Macs, and every later version is included.
    • Entering your key activates this Mac once, online. After that the app keeps working with no connection.
    • If it is not for you, email support@pocketweb.tools within 14 days of buying and we refund the full amount.
    Release

  2. Edit: cut a video by editing its transcript

    • New Edit screen. Drop a video or audio file of any length and size, or record from the microphone or camera, then delete words from the transcript to cut them from the media. Undo brings any cut back.
    • Remove fillers and Shorten pauses each take one click. Parakeet TDT v3 writes the words, and a small word aligner (wav2vec2, Apache 2.0, 90 MB, English) times each one and catches the ums and uhs between them. On a 90-second meeting recording, Remove fillers cut 18 of 19 fillers where the browser version cuts 13; it sometimes clips a short word next to a filler, and one undo restores it.
    • Export MP4 with captions burned in, in the same four styles as Clip, M4A or WAV audio, or the transcript as SRT, WebVTT or plain text. Exports run on AVFoundation, so no browser tab limits the file size.
    • Opens MP4, MOV, M4V, M4A, MP3, WAV, AIFF, CAF, AAC and FLAC. Editing needs Parakeet TDT v3, so it works in Parakeet's 25 European languages. WebM and MKV are not supported because macOS has no reader for them, and MP3 export is not in this version.
    • Every edit is saved to the history and reopens with its cuts. Recordings made in the editor are kept on this Mac with your 10 most recent edits. The pause detector (Silero VAD, MIT) ships inside the app; the aligner is a one-time download, offered in the editor when you edit English.
    Pre-release

  3. The word counter's keyword panel, on the app's own engine

    • The word counter now has its keyword panel, which was the last thing the app was missing next to the website. Type or paste anything and the panel ranks the phrases the text is actually about, by meaning rather than by how often a word repeats.
    • It runs on a small model built into the app, on the same engine the chat and transcription screens use. Nothing is downloaded the first time you open it, there is no progress bar to sit through, and it adds nothing to the short list of things the app talks to.
    • The website version downloads a keyword model into your browser to do the same job. The app already had an engine for this, so it uses that instead.
    • Same phrases, same one, two and three word tabs, same counts and densities as the web tool: both sides now share one piece of ranking code.
    • The GPT token count is now exact in the app, and it was not before: it runs OpenAI's own tokenizer, built into the app, instead of the browser code the website has to use. It reads the same number OpenAI bills, on a long document, in about a millisecond.
    Pre-release

  4. Every tool from the website is now in the app

    • The fifteen everyday utilities from pocketweb.tools now live in the app under Utilities, so you do not have to open a browser tab for the small jobs: word counter, currency converter, timezone converter, math notepad, image compressor, social media image resizer, JSON formatter, Base64 encoder, URL encoder, case converter, UUID generator, password generator, hash generator, lorem ipsum generator and noise generator.
    • They are the same tools, not rebuilds: the app and the website share one component tree, so a fix on either side lands on both.
    • Saving is native. Compressed or resized images open a save panel, and a whole batch asks once for a folder instead of firing one download per file the way a browser has to.
    • Two of them need the internet for today's exchange rates, and only for that: the currency converter and currency lines in the math notepad. The privacy section on the product page lists that call alongside the other three. Every other utility works with Wi-Fi off.
    • The word counter shipped without its keyword panel here. It arrived on 2026-09-16, running on the engine the app already loads.
    Pre-release

  5. Clip: long videos into short captioned clips

    • New Clip screen. Drop a talking video of any length, or a whole folder, and a chat model on your Mac reads the transcript and picks the self-contained moments, each with a title and the reason it works. Adjust any clip in the same editor as the web tool, or add your own at the playhead.
    • Transcription runs on Parakeet TDT v3 with word timestamps, or Whisper Large v3 Turbo for 99 languages, and the picks come from a chat model you already have, Qwen3.5 9B by default. A 25-minute talk was transcribed and picked in about two and a half minutes on an M1 Pro.
    • Exports run on Apple's AVFoundation and the VideoToolbox hardware encoder: about 190 frames per second at 1080p on an M1 Pro, so a 30-second captioned 9:16 clip takes about five seconds.
    • 9:16, 1:1 and 4:5 with a blurred background, a centre crop, or a crop that follows a face shot by shot using Apple's Vision framework. Four caption styles in five bundled typefaces (Anton, Bebas Neue, Montserrat, Poppins and Inter, all SIL Open Font License), a pause trim that listens to the audio so it never cuts through a word, and an SRT or WebVTT file beside every clip.
    • MP4, MOV and M4V in; H.264 or HEVC MP4 out, up to 4K from a 4K source. MKV and WebM are not supported yet because macOS has no reader for them.
    Pre-release

  6. Isolate: background removal and click-to-pick at full resolution

    • New Isolate screen with two modes. Everything in front removes the background from every photo you drop, or a whole folder, at the photo's full size. Pick a subject reads one photo once, then every click or box returns a mask you can save; add points to include or exclude parts.
    • Two engines in both modes. Instant uses Apple's Vision framework built into macOS: nothing to download, about a second a photo, a hard edge. The downloadable tiers give real hair, fur and glass: BiRefNet lite (MIT, 82 MB) and BiRefNet (MIT, 407 MB) for clean cutouts, BiRefNet matting (MIT, 407 MB) for soft edges, SAM 2.1 Small (Meta, Apache 2.0, 184 MB) and EfficientViT-SAM XL1 (Apache 2.0, 814 MB) for picking.
    • Full-resolution edges: ViTMatte (MIT, 113 MB) refines the edge of any mask over the original pixels, tile by tile, so a 24 megapixel photo keeps whisker-level detail instead of a 1024 pixel cutout scaled up. On by default when installed; about a second per photo on an M1 Pro, half a second for the lite tier, a few hundredths of a second per click when picking.
    • Opens what macOS opens, including HEIC and camera RAW. Saves PNG, WebP or TIFF with transparency, JPEG over a colour of your choice, or the mask alone as a greyscale image. Colour profiles and EXIF travel into PNG and JPEG results.
    • Fourth engine: Apple's Core ML framework runs the BiRefNet bundles directly on the GPU; the pickers and ViTMatte run on ONNX Runtime through Core ML. Their one-time Core ML compile happens right after the download, not on the first click. Every license text ships inside the app.
    Pre-release

  7. Speak: text to speech and voice cloning in ten languages

    • New Speak screen. Paste or type any amount of text and get it read aloud in a built-in voice or in a voice cloned from a few seconds of audio: a Voice Memo is enough. Play it in the app, then save it as M4A or WAV. Blank lines become pauses between paragraphs.
    • Ten languages: English, Chinese, Japanese, Korean, German, French, Spanish, Italian, Portuguese and Russian, from one model, with no character limit.
    • Voice model: Qwen3-TTS 1.7B (Qwen, Apache 2.0, 1.5 GB), running on the same llama.cpp engine as the chat models, on the GPU. About 1.4 seconds of speech per second on an M1 Pro, so an hour of narration takes about forty minutes.
    • Saved voices stay on this Mac, and cloning asks you to confirm you have the speaker's permission. The license text ships inside the app.
    Pre-release

  8. Extract: scans, screenshots and PDFs to text, markdown or a searchable PDF

    • New Extract screen. Drop PDFs, scans or screenshots, or a whole folder, and get plain text, markdown, or a searchable PDF with an invisible text layer placed on the printed words. Pages that already have text are kept as they are instead of being re-read.
    • Two engines, one switch. Instant uses Apple's Vision framework built into macOS: about a second a page, nothing to download, in the languages your version of macOS reads. Document model downloads a small vision model that reads layout: headings, tables as tables, formulas as LaTeX, and scripts Vision cannot read.
    • Document models: GLM-OCR (Z.ai, MIT, 1.4 GB, about 7 seconds a page on an M1 Pro) as the default, dots.ocr (rednote, MIT, 3.2 GB, 100 languages) for multilingual documents. Both run on the same llama.cpp engine as the chat models, on the GPU.
    • Send to Chat hands the extracted text to the chat screen so you can ask the 9B or 27B model about the document. License texts for both models ship inside the app.
    Pre-release

  9. Upscale: full resolution, whole folders, HEIC and RAW

    • New Upscale screen. Drop photos or a folder, pick 2x or 4x, and get the result at full resolution: the web tool shrinks every input to 512 pixels first, the app runs the real file, tile by tile, on the GPU.
    • Opens what macOS opens: JPEG, PNG, TIFF, WebP, HEIC straight from an iPhone, camera RAW. The colour profile and the EXIF data travel into the result.
    • Batch: a whole folder runs as one queue with per-file progress, and the output lands next to the originals or in a folder you choose.
    • Third engine: ONNX Runtime with Apple's Core ML, statically linked. Models: RealPLKSR 4x (the default for web and phone photos) and 2x, Real-ESRGAN x4plus for large photos, Real-ESRGAN anime. Every one licensed for commercial work; the license texts ship inside the app.
    Pre-release

  10. Transcribe: files, live audio, system audio, speaker labels

    • New Transcribe screen. Drop any audio or video file this Mac can play, with no size limit, and get a transcript with timestamps, a segmented or plain view, and export as text, SRT, WebVTT or JSON.
    • Live mode transcribes the microphone, the audio the Mac is playing (a call, a video, any app) or both, as it happens. System audio uses a Core Audio process tap; macOS asks once for permission.
    • Speaker labels: NVIDIA's Sortformer diarizer marks who spoke when, up to four speakers, on files and at the end of a live session.
    • Second engine: transcribe.cpp on Metal next to llama.cpp. Speech models: Parakeet TDT v3 (25 languages, word timestamps, the default), Whisper Large v3 Turbo (99 languages, translate to English), Nemotron 3.5 Streaming (live), Sortformer (speakers).
    • Every model license the app redistributes under now ships inside the app, listed on the License screen.
    Pre-release

  11. Qwen3.5 9B becomes the recommended model for 16 GB Macs

    • Bake-off on a 16 GB M1 Pro across the three 16 GB-class models: Qwen3.5 9B answered fastest (19 tokens per second), scored best on the test prompts and is the smallest download of the three, so it is now the first pick for 16 GB Macs. Gemma 4 12B stays in the lineup as the alternative.
    • Gemma 4 works: the app now understands its chat format, which the bundled llama.cpp build did not.
    • Gemma 4 no longer runs out of GPU memory at a 16k context on 16 GB Macs (sliding-window layers use a window-sized cache: 0.7 GB instead of 5.4 GB).
    • Follow-up turns in a thread reuse the cached prompt on every model, so the second message in a long chat starts streaming in well under a second instead of re-reading the whole conversation.
    Pre-release

  12. Markdown replies, prompt caching, turn actions

    • Replies render as Markdown with syntax-highlighted code blocks and tables.
    • Copy and regenerate on every reply; stopping a reply keeps the text streamed so far.
    • A context meter shows how much of the model's window the thread has used and warns before it overflows.
    • Prompt caching across turns: the model keeps the conversation in memory between messages instead of re-processing it.
    Pre-release

  13. First build: local chat with six models

    • Chat screen with threads saved on your Mac, and a Models screen that downloads, verifies and loads models on demand.
    • Lineup: Qwen3.8 27B (flagship, 32 GB Macs), Qwen3.6 35B‑A3B (fast mixture-of-experts, 32 GB), Gemma 4 12B and Qwen3.5 9B (16 GB), Bonsai 27B 1‑bit and Qwen3.5 2B (8 GB). All Apache 2.0.
    • Engine: llama.cpp on Metal with every layer on the GPU, streaming tokens to the interface as they are generated.
    • Downloads resume after an interruption and are checked against the publisher's SHA-256 before a model can load.
    Pre-release
Buy for $99$49

Already have a key? Download the app. It asks for the key the first time it opens.

14-day money-back guarantee. Your license key is shown right after checkout and stays on your purchase page; one key works on up to 3 Macs.