Changelog

Product updates

A focused feed of meaningful product changes and polish improvements.

v0.4.16

2026-10-31

Arrange the active window from LookSpeak's menu bar using halves, 4/6/8-zone grids, maximize, restore, and display-switching commands.
Settings → Shortcuts now includes optional window-arrangement bindings with destination previews and conflict checks. All new shortcuts start unassigned.
Window arrangement works independently of dictation and gaze calibration, preserves per-window restore history, and reports application size limits.
Smart model defaults adapt CPU threads and shared transcription/cleanup retention to your Mac: one minute on 8 GB, two minutes on 16–18 GB, and five minutes on 24 GB or more.
S1 Polish dictation is automatically enabled for new configurations above 16 GB RAM, with automatic downloads, manual overrides, memory-pressure protection, and clear Settings-only availability status. Existing explicit choices are preserved.
Requires macOS Sequoia 15.5 or later on a Mac with Apple silicon (M1 or later).

v0.4.15

2026-10-31

The HUD now keeps Refining visible throughout experimental cleanup and pasting, then shows Inserted after successful delivery.
Polish dictation now uses S1-mini by Superwhisper exclusively, with a compact Models setting, blue toggle, and clear download progress.
Cleanup remains experimental and off by default. Existing S1 opt-ins are preserved; older model selections require enabling S1 before its first download.
Requires macOS Sequoia 15.5 or later on a Mac with Apple silicon (M1 or later).

v0.4.14

2026-10-31

Fixes experimental cleanup preparation in the installed app, including S1-mini by Superwhisper, by launching the correct bundled worker.
Hold-to-speak now preserves queued and briefly delayed microphone audio when the shortcut is released, helping protect the final words.
Adds regression coverage for rapid holds, delayed microphone input, cancellation, and the packaged cleanup preparation path. Cleanup remains optional and off by default.
Requires macOS Sequoia 15.5 or later on a Mac with Apple silicon (M1 or later).

v0.4.13

2026-10-27

VS Code dictation now uses the same safe paste path as Copilot, preserving existing text, manual line breaks, and the live cursor position.
Experimental post-transcription cleanup now offers S1-mini by Superwhisper as an optional English-only model alongside Qwen.
The Models page shows clearer cleanup model download, preparation, and readiness states. Cleanup stays off by default and model files download only when enabled.
Requires macOS Sequoia 15.5 or later on a Mac with Apple silicon (M1 or later).

v0.4.12

2026-10-25

Resumed dictation preserves existing text and manual spaces or line breaks, including in Copilot and Codex.
Models now offers Post-transcription cleanup (experimental), off by default, with a separate local model downloaded only when enabled.
Standard cleanup better preserves complete sentences, questions, names, and technical text without requiring the experimental model.
Requires macOS Sequoia 15.5 or later on a Mac with Apple silicon (M1 or later).

v0.4.11

2026-08-24

Automatic updates now open and install normally instead of failing with a generic updater error.
This release corrects the Sparkle helper signing used by macOS security checks.
If you are on 0.4.9 or 0.4.10, install 0.4.11 manually once; automatic updates will resume afterward.

v0.4.10

2026-08-24

Trusted recipient and window context now preserves names more reliably in greetings and direct address without broad capitalization changes.
Email greetings now use a dedicated greeting line and blank line before the message, while chat and terminal greetings remain compact.
GrokBot dictation now inserts once instead of duplicating text, and Microsoft 365 comments no longer inherit stale leading spaces.

v0.4.9

2026-08-08

LookSpeak now binds paid access to this Mac with signed, device-specific authorization instead of reusable local credentials.
Linked-device access now refreshes through one-time signed challenges and fails safely if entitlement data is altered or replayed.
Billing sessions now use stronger host-only cookie protections while existing linked Macs can migrate without losing access.

v0.4.8

2026-08-08

Transcript feedback now includes simple controls for line breaks, paragraphs, bullets, and numbered lists while preserving the exact corrected layout.
Reviewed feedback now distinguishes wording changes from formatting-only corrections so each signal can be used accurately.
Feedback retries across LookSpeak API domains now reuse one submission ID, preventing duplicate reports during network fallback.

v0.4.7

2026-08-08

Camera and microphone discovery now runs outside the main interface thread, preventing device enumeration from delaying onboarding and settings.
LookSpeak now checks for updates automatically and can download them in the background for installation when the app quits.
A new Help & Updates control lets each user keep automatic downloads enabled or choose to approve updates manually.

v0.4.6

2026-08-07

Reviewed transcript corrections can now teach bounded contextual product names while continuing to reject ordinary wording and formatting edits.
Trusted names in thank-you messages now receive the intended capitalization and direct-address punctuation without broad sentence rewrites.
Context-aware formatting now keeps generic uses of connect lowercase while preserving Amazon Connect product references.

v0.4.5

2026-08-05

Reviewed transcript corrections now teach bounded personal vocabulary for names and technical terms while ignoring ordinary wording and formatting edits.
Transcription now preserves dotted technical identifiers and formats clock times with a.m. or p.m. without introducing false sentence breaks.
Context-aware formatting restores supported technical terms only when matching local evidence is present and prevents common words from becoming global capitalization rules.

v0.4.4

2026-08-05

Names from trusted window and recipient context now retain their intended capitalization without changing unrelated words.
Billing state checks now run outside the main UI thread, reducing avoidable interface stalls while preserving account refresh behavior.

v0.4.3

2026-08-04

Names from the current window and recipient context now guide on-device transcription and capitalization while protecting ordinary words and structured text.
Reviewed name corrections can now become bounded app- and context-aware vocabulary entries without rewriting unrelated prose.
Transcript history now records recognition context and applied corrections for clearer local feedback diagnostics.

v0.4.2

2026-08-04

Context-aware transcription now handles names more accurately and can learn clear name corrections from transcript feedback.
Calibration now remains valid when the same display changes proportional macOS scaling, while still detecting rotation and material aspect-ratio changes.
Account refreshes now back off during connectivity and service outages and recover cleanly without repeated error reporting.
Release builds now retain and upload Swift debug symbols for more useful crash diagnostics.

v0.4.1

2026-07-31

Personal vocabulary now applies corrected name capitalization sooner while continuing to ignore ordinary sentence-start capitalization.
Transcription now fixes accidental all-caps common words while preserving acronyms and identifiers.
Models settings now distinguish model size from active memory and explain the storage and memory impact of transcription acceleration.

v0.4.0

2026-07-30

On-device transcription acceleration is now included and verified as a required part of every customer build.
LookSpeak can adopt accelerated transcription automatically after its local model asset is ready while preserving the validated local fallback.

v0.3.9

2026-07-30

LookSpeak can now use verified on-device transcription acceleration when it is available, while safely retaining the existing local fallback.
Transcription performance and model memory retention are now easier to tune in Settings.

v0.3.8

2026-07-29

Live transcript previews now start promptly, while a centered listening state remains visible only until interim words arrive.

v0.3.7

2026-07-29

Calibration now safely reads non-ASCII subprocess output, preserving diagnostics instead of failing during background decoding.

v0.3.6

2026-07-28

Bulleted dictation now uses native rich-text lists in supported editors while preserving the exact plain-text fallback elsewhere.
Text insertion now recognizes when focus moves to a different field, preventing stale leading spaces or list context from carrying across inputs.

v0.3.5

2026-07-27

Fine-grained calibration now checks accuracy across smaller screen regions and keeps refinements only when independent validation improves the result.
Per-display gaze targeting now retains the strongest validated blend of coordinate and zone predictions while avoiding unnecessary recalibration for display-position changes.
The transcript HUD is slightly denser, with interim and status text centered evenly from top to bottom.

v0.3.4

2026-07-27

Personal vocabulary now learns short corrections with local context while staying bounded, editable, and entirely on device.
Spoken formatting better distinguishes explicit commands from ordinary phrases and follows the live cursor and active list context.
Hold-to-speak transcript previews can appear sooner on supported runtimes while final transcription remains the quality authority.

v0.3.3

2026-07-23

Fixes a startup crash caused by the hold-to-speak keyboard listener on current macOS releases.
Restores complete desktop crash reporting so future failures can reach Sentry.
Live transcript previews now take priority over the centered listening state.

v0.3.2

2026-07-22

Long-form dictation now handles pauses and final text more reliably.
Microphone and transcription workers now recover more safely without leaving stale background work behind.
Listening and inserted-text feedback in the transcript HUD is clearer and disappears more promptly.

v0.3.1

2026-07-20

Purchasing, account linking, subscription checks, and billing callbacks now use lookspeakapp.com for better compatibility on managed company networks.
The canonical website, downloads, updates, and optional analytics remain on lookspeak.ai.

v0.3.0

2026-07-19

Personal vocabulary now learns corrected spellings and joined terms, persists across restarts, and supports fast batch entry in Settings.
On-demand transcription finishes faster without sacrificing long-form quality, and decoder context no longer leaks vocabulary labels into text.
Gaze activation responds faster after speech pauses, while adaptive calibration is clearer and more resilient across multiple displays.
Transcripts, shortcuts, and the Home experience are easier to scan with denser, more focused layouts.

Multi-display support

Product milestone

2026-05-26

LookSpeak gained per-display calibration, so gaze targeting could move reliably between a MacBook and connected monitors.
A dedicated Monitors setup made it clearer to add, review, and recalibrate every display from Settings.
Onboarding began guiding people through multi-display setup before they started working across a larger desk.

v0.2.9

2026-03-31

Text insertion now recovers focus more reliably in apps like Cursor, Codex, and browser-based editors.
Focused text context can now be benchmarked end to end, with a new evaluation harness and release docs for transcription testing.
Cursor-aware context handling is steadier, with safer fallback behavior when editable targets are harder to reach.

v0.2.8

2026-03-31

Gaze targeting and dictation now feel steadier, with faster activation and more reliable text insertion.
Model setup is cleaner, with a simpler local default and a tighter memory-retention control.
Billing and entitlement refreshes fail more quietly when the network is unstable, while real server errors are easier to diagnose.

v0.2.7

2026-03-28

Long-form dictation now handles pauses more cleanly, with better sentence endings and fewer stray capitals.
Model setup feels cleaner, and browser chat inputs like Chrome avoid duplicate inserts.

v0.2.6

2026-03-23

Home onboarding now explains first use more clearly inside Settings.
The home preview and guidance layout feel tighter and easier to scan.

v0.2.5

2026-03-22

Best is now the default transcription model for new installs.
Help & Updates stays shorter and easier to scan.

v0.2.3

2026-03-22

Help & Updates now uses shorter copy so key actions stay easier to scan.
Your current version now shows the release date instead of an internal build number.
What's new stays compact, so the changelog and support links remain visible.

v0.2.2

2026-03-22

First-run onboarding is steadier again, with a calmer Dock and permission flow.
Calibration now recovers more gracefully, with clearer guidance and a reliable retry path.
Plan & Limits, Get Plus, and update checks fail more cleanly when something goes wrong.

v0.2.1

2026-03-17

The installer is back to the polished drag-to-Applications experience, including the branded DMG background and layout.
First-run onboarding and the post-setup settings handoff are steadier, with less visual churn while permissions and setup complete.
Settings headers now sit a touch closer to the title bar for a tighter, cleaner layout.

v0.2.0

2026-03-17

First-run setup now uses the main LookSpeak app as a proper foreground owner for camera and microphone prompts during onboarding.
Microphone and camera setup now share the same native AVFoundation permission path for more predictable macOS prompt behavior.
Startup keeps the faster onboarding handoff from 0.1.9 while leaving the heavier background runtime until setup is finished.

v0.1.9

2026-03-17

First launch now reaches onboarding more quickly on fresh Macs by delaying the heavier background runtime until setup is complete.
Camera and microphone setup are now coordinated by the main LookSpeak app during onboarding, while calibration and model prefetch still continue in the background.
The installer now stages a deterministic Applications alias with an explicit folder icon for a steadier drag-to-install experience.

v0.1.8

2026-03-16

Camera and microphone setup now tracks macOS permission results more accurately during first launch, especially when prompts take a moment to settle.
The onboarding window now launches from the packaged helper more reliably on fresh Macs instead of depending on a temporary copy.
The installer’s Applications target has been adjusted again for more consistent rendering in the drag-to-install window.

v0.1.7

2026-03-16

Camera and microphone setup now keeps macOS permission state more reliably during first launch, especially when permission prompts take a moment to settle.
The installer renders more consistently across Macs, including a clearer Applications target in the drag-to-install window.
Onboarding handoff and recovery have been tightened again after fresh-Mac testing.

v0.1.6

2026-03-16

Camera and microphone permissions now prompt correctly during first-time setup on fresh Macs.
The installer’s drag-to-Applications target has been adjusted for more consistent rendering across Macs.
Onboarding reliability has been tightened again after real-world first-launch testing.

v0.1.5

2026-03-16

Permissions now start more reliably on fresh Macs, especially for camera and microphone access during onboarding.
The onboarding window stays more stable during first launch, so setup is less likely to get interrupted.
Installing LookSpeak is more consistent across Macs, including a more reliable drag-to-Applications experience in the installer.

v0.1.4

2026-03-16

Onboarding now opens more reliably on fresh Macs, with a steadier permission and setup flow.
Installing LookSpeak is more consistent across Macs, including a more reliable drag-to-Applications experience in the installer.
First launch is smoother overall, with better recovery when setup doesn’t go to plan.

v0.1.3

2026-03-16

First-run onboarding is faster and more reliable, including calibration, permissions, and the finish-setup handoff.
Website downloads now start directly from the live site and guide new installs more clearly.
Local transcription defaults to the Balanced model, with Best, Balanced, and Fast labels in Settings.

v0.1.2

2026-03-15

Calibration became more stable in the packaged app, with better fullscreen handling and smoother recovery when setup got interrupted.
The app download was made significantly smaller and easier to install, with a cleaner DMG and website install flow.
The onboarding finish flow was tightened so new users could get from install to first use more consistently.

v0.1.1

2026-03-13

First-run onboarding became more reliable, especially around permissions, calibration, and getting into the app cleanly after setup.
Settings were polished to make key actions easier to find, including updates, plan details, and model selection.
The website and download experience were refined so installing LookSpeak felt clearer for new testers.

v0.1.0

2026-03-12

LookSpeak reached its first public beta on macOS, including local gaze-to-text, calibration, window targeting, and the first onboarding flow.
Plan & Limits, Plus activation, and the website download path were introduced so early testers could install and use the full product end to end.
Privacy messaging, help surfaces, and the overall LookSpeak brand were brought into the app and website.