Window control
Point anywhere inside a window, hold a chord, and move or resize it directly. No title-bar grab or corner targeting is required, and snapping stays predictable across displays.
macOS 26+ · Apple Silicon · Public Release
Mattouch
Move and resize the window under your cursor with a modifier chord. Dictate in your own words, find a window by its contents, or add local vision and guided automation. Choose the smallest bundle that fits your work. No analytics, automatic reporting, or automatic update checks.
Current Mattouch
macOS 26+
- Current Version
- 1.2.0
- Bundles
- 5 additive tiers
- Minimal Version
- 3.6 MB zip
- Maximal Version
- 17 MB zip
Downloads
Current Mattouch
A more personal dictation workflow with microphone setup, real-hotkey practice, optional vocabulary and stronger capture, clipboard and model ownership.
Mattouch (Window)
Move and resize windows with hotkeys plus side-fill snapping
Complete
Core window control is the stable baseline and expected to be correct on supported systems.
3.6 MB zip / 4.5 MB dmg
Mattouch (+ASR)
Local hotkey dictation, microphone setup and optional vocabulary controls
Stable
Dictation workflows are usable; model behavior can still vary by selected speech runtime.
6.6 MB zip / 7.8 MB dmg
Mattouch (+OCR)
Search open windows with automated OCR of all window visible content at once
Usable
Window search is functional, with accuracy affected by screen-recording permission and app content.
6.7 MB zip / 7.9 MB dmg
Application features
What Mattouch does
Fast window control at the core. Add dictation, search and local intelligence as your work needs them.
Local by design
No analytics, automatic reporting, or update checks. Network access is used for model downloads and APIs you explicitly configure. Support details are copied locally only when you choose.
Controlled automation
The full tier can plan actions from a plain-language goal, while real clicks and typing stay behind app allowlists and confirmations.
Adaptive dictation
Use Apple Speech, Whisper, Parakeet or Nemotron with explicit model and language readiness. Compare installed engines with Hyper, review disagreements, and see provisional text where the selected engine supports preview.
A microphone you can trust
Choose System Default or a specific microphone without changing the Mac's input setting. Run a five-second level test, then practice with your real dictation hotkey in Settings before using another app.
Your vocabulary, your formatting
Add optional word replacements, reusable snippets and spoken punctuation. Preview drafts before applying them, import or export vocabulary, and choose Preserve, Sentence or Inline text styles globally or per app.
Careful delivery and recovery
Default cleanup preserves code, Markdown and multiline text. Dictation protects rich clipboard contents and newer copies, and cancelled results cannot arrive later. Recent transcripts can be recovered temporarily without a saved transcript history.
Reviewable transcript rewrites
Optionally rewrite a transcript with a local or configured API model, inspect the before-and-after diff, adjust a preset or one-run instruction, regenerate, cancel, edit, or keep the original.
Extensible local models
Browse compatible Hugging Face MLX text models, inspect model metadata and disk requirements, and follow live byte-level download progress from one shared model lifecycle.
New in 1.2 · +ASR and above
Make dictation feel like your Mac
Set up once in Settings → Dictation, then use your hotkey where you write.
1. Check your input
Pick a microphone and test its levels. Check the selected engine's readiness and supported language before recording.
2. Try your real hotkey
Start practice and dictate into the Settings field. Confirm microphone and Accessibility permissions, then choose paste or copy-only delivery.
3. Make the words yours
Add your names, phrases and snippets. Preview optional punctuation and text styles before applying them globally or for one app.
Release bundles
Start small. Add what you need.
Each bundle includes the tiers before it. Choose Window for gestures alone, +ASR for dictation, +OCR for text search, +VLM for local vision, or +Agent for guided actions.
| Bundle | Best for | Status | Capabilities | Dependencies |
|---|---|---|---|---|
| Mattouch (Window) Core utility |
Users who only want fast chord-driven window control. |
Complete
Core window control is the stable baseline and expected to be correct on supported systems.
|
Move and resize windows with hotkeys plus side-fill snapping
|
No third-party runtime dependencies |
| Mattouch (+ASR) Adds dictation |
Users who want window control plus local speech-to-text. |
Stable
Dictation workflows are usable; model behavior can still vary by selected speech runtime.
|
Local hotkey dictation, microphone setup and optional vocabulary controls
|
Adds WhisperKit and FluidAudio |
| Mattouch (+OCR) Adds all-window search |
Users who want to find and raise windows by title, text, or visible content. |
Usable
Window search is functional, with accuracy affected by screen-recording permission and app content.
|
Search open windows with automated OCR of all window visible content at once
|
ASR dependencies plus Apple Vision/NaturalLanguage system frameworks |
| Mattouch (+VLM) Adds local vision |
Users who want richer visual matching without hosted inference. |
Preview
Local vision improves matching, while model download, runtime, and ranking behavior are still being tuned.
|
Adds local AI image input models to window search features
|
Adds MLX, mlx-swift-lm, swift-huggingface, and swift-transformers |
| Mattouch (+Agent) Full Full feature set |
Users who want local goal planning with control over any real action. |
Beta
Agent planning is actively evolving and real actions remain guarded by allowlists and confirmations.
|
Let the local AI models move your mouse and click things and type things too
|
Full local inference stack |
Release notes
What's new in 1.2.0
A more personal dictation workflow with microphone setup, real-hotkey practice, optional vocabulary and stronger capture, clipboard and model ownership.
Choose, test and practice your microphone
Select a stable input without changing the system default, check its levels in a cancellable five-second test, and practice your existing hotkey in Settings. A missing selected microphone is reported explicitly. The microphone test does not save audio.
Vocabulary and snippets with a preview
Configure optional whole-word replacements, multiline snippets, spoken punctuation and Preserve, Sentence or Inline styles. Preview and Apply keep draft edits separate from active dictation; JSON import/export and per-app overrides make the setup reusable.
Honest engine and language readiness
Settings, model inventory and dictation share selected-component readiness. Apple assets are locale-specific, Whisper accepts a language choice, Parakeet detects language and Nemotron is English-only. Hyper uses installed, ready engines without incidental downloads.
Preserve text and protect your clipboard
Cleanup removes known noise labels while preserving ordinary parentheses, code, Markdown and line breaks. Coordinated paste transactions preserve rich clipboard types, protect newer copies and resolve Command-V using the active keyboard layout.
Cancellation and input interruption recovery
Cancelling dictation prevents late results from being pasted. Models in use are protected from deletion, and microphone interruption recovery retains captured speech and tells you when a gap occurred.
Smoother previews and model readiness
Provisional text settles as you speak while final recognition uses the complete recording. Optional prewarming prepares already installed speech models without opening the microphone or downloading anything, and idle models unload when they are no longer needed.
No stray words from empty input
When a device sends completely silent digital audio, dictation finishes without inserting stray words, including in Hyper. Background noise still follows normal speech recognition.
Download details and checksums
Compare a downloaded file with its SHA-256 checksum below.
| Artifact | Bytes | SHA-256 |
|---|---|---|
| Mattouch-window-1.2.0.zip | 3,584,003 | b9f6e1fbe06a97bbb02158523cb29e322da5c9ae86201afc941167a175c5b6d4 |
| Mattouch-window-1.2.0.dmg | 4,494,374 | 7f55ac3dc0da2f475b166df82574d00ab91ae27b93bf15d1e855c07b575acc28 |
| Mattouch-asr-1.2.0.zip | 6,583,076 | c6fcc2d45ea7ff5c2ed15efa19b3f7baad47fff045cde8a9aaf857ad618b2f41 |
| Mattouch-asr-1.2.0.dmg | 7,792,723 | c51ec5f516dfcac7a90adf3b750236f0a45c0e64e7931de245bc19fa14c1a51e |
| Mattouch-ocr-1.2.0.zip | 6,699,908 | 2432ca9e5376f95414ab327b7c7282171113aea3652effc2038c6f4b316491e8 |
| Mattouch-ocr-1.2.0.dmg | 7,922,352 | 4f9cf1399ba5da9ab891ef2123f272c10acb8b5f2c7963fc5bcf3b5830e0eebf |
| Mattouch-vlm-1.2.0.zip | 16,273,868 | 58c1d03c6abebc61a1c2afb8537faab55a02b97aa6fc9151224c6e771081b2e0 |
| Mattouch-vlm-1.2.0.dmg | 18,783,012 | 09653bb1370ccbfceda7c2abf478f7a2ff6cc097abeff6b2b9419b96e3d751b6 |
| Mattouch-agent-1.2.0.zip | 17,380,180 | 625d74188f2f5e851599893a0911960e1eea2ae2b2011ec9bafa3dacc06bd970 |
| Mattouch-agent-1.2.0.dmg | 19,981,569 | d35841f5e1c067a5d007faa9ea31096426ef778715235186173f30c521f12719 |
Privacy
Local means local
Mattouch does not phone home. There is no telemetry, analytics, crash reporting, remote reporting, or automatic update check. If you want a newer version, come back here and compare the release archive yourself.
On-device recognition
Local speech and AI engines process your content on your Mac. You choose which external models to download.
Explicit API choices
Optional rewriting can use an API you configure. Its provider receives the content you send through that workflow.
Your updates, your support report
No automatic update requests or reporting. Copy for support creates an allowlisted clipboard report only when you ask; recordings and transcripts are excluded.
Support
Feedback and help
Questions, bugs, release feedback, and setup issues are welcome by email or in the community Discord.
Release history
Release archive
Every published version appears here, including the current Mattouch, with generated links for every bundle artifact.
| Version | Date | Summary | Release notes | Downloads |
|---|---|---|---|---|
| 1.2.0 Current | 2026-10-08 | A more personal dictation workflow with microphone setup, real-hotkey practice, optional vocabulary and stronger capture, clipboard and model ownership. |
|
|
| 1.1.3 | 2026-09-15 | Improves Bluetooth dictation, volume restoration, recording feedback, and local support reports. |
|
|
| 1.1.2 | 2026-09-04 | Makes hotkeys, dictation, model downloads, rewriting and status overlays work more consistently together. |
|
|
| 1.1.1 | 2026-08-29 | Makes grab-anywhere window movement and resizing responsive in browser and other asynchronously rendered windows. |
|
|
| 1.1.0 | 2026-08-23 | A major dictation, transcript-rewrite, model-management, and grab-anywhere window-control upgrade. |
|
|
| 1.0.0 | 2026-06-07 | Initial tiered release with five dependency-stripped macOS bundles. |
|