Changelog

Resonant now checks whether your words actually appear in the text field. Before, small formatting differences in web apps like Gmail, ChatGPT, WhatsApp Web and LinkedIn made many successful pastes look unconfirmed. Each of those put a backup copy on your clipboard and showed a "⌘V if needed" notice.

When a paste really can't be confirmed, the backup copy now holds your clipboard for 1 second instead of 10. After that, press ⌃⌘V to paste your last dictation.

When a paste can't be confirmed, Resonant still puts your dictation on the clipboard so you can press ⌘V. That copy is now temporary. A ring next to your mouse pointer counts down 10 seconds, and then whatever you had copied before comes back. If you press ⌘V first, your clipboard comes back right after the paste.

When Resonant pastes or sends your dictation, a small check now appears next to your mouse pointer. The notch closes right away instead of staying open to say "Pasted!".

This update fixes a crash in 0.1.96 that closed Resonant when you opened Dictation settings.

Resonant now keeps the words you say right as you press the shortcut, and every dictation is saved before it's pasted. If a paste doesn't land, Paste last dictation (⌃⌘V) puts it right back.

This release also fixes Instant-on presses that failed to start, keeps numbers and units the way you said them, and removes a set of stalls caused by background work.

Resonant now starts preparing the microphone the moment you press the shortcut, and live preview shares the recording and the loaded speech model instead of running its own.

Cleanup now keeps what you meant: questions stay questions, numbers stay exact, and a "not" is never dropped. Resonant also remembers which microphone you prefer and falls back sensibly when it disappears.

Under the hood, stalled transcriptions now time out and can be retried, speech-model downloads resume where they left off, and meetings keep recording smoothly on a busy or hot Mac.

Signing in with Google no longer fails if it takes you longer than 30 seconds — picking an account, typing a password, or completing two-factor authentication now has plenty of time.

Fixes an issue in 0.1.90 and 0.1.91 where one interrupted recording could block every new dictation, even after restarting. Your next recording now always goes ahead; any recording that couldn't be recovered automatically is set aside on disk rather than deleted.

Resonant now learns from you: correct a word once and it adapts, learns your formatting preferences from how you edit, and recognizes the names it sees on your screen — all on this Mac, all governed by your privacy settings.

This release also makes Resonant genuinely usable with VoiceOver and keyboard-only setups, and first-time setup is more resilient: the speech-model download shows honest progress, recovers from network hiccups on its own, and never leaves the app stuck.

Pasting feels noticeably faster, especially in webmail like Gmail and Outlook, and your clipboard is preserved when you paste there. Cleaned transcripts also repair fragmented words and trim abandoned half-phrases more reliably.

Meeting transcripts now capture speech reliably.

Meetings now appear alongside your dictations in the daily journal — merged chronologically with dedicated icons, a teal accent, and duration stats. Your meeting history, notes, and summaries stay in sync across devices.

This release also brings Zen browser support, smarter paste spacing that understands sentence boundaries, and RTF clipboard support so links paste correctly in rich-text editors like Apple Mail and Pages.

Meeting recordings are now more reliable. A stop button in the scratchpad gives you direct control, and if your mic drops out mid-meeting the recording auto-recovers.

Resonant now supports 30 languages including Swedish, Italian, French, German, and more. Pick your language in Settings or leave it on auto-detect.

Meeting recording is now manual-only — start it from Settings or the notch menu, never automatically. Auto-detect and meeting app toggles are gone in favor of a simpler, more predictable experience.

Meeting recording defaults to manual, with a reworked Meetings settings panel. No more surprise recordings.

Your workspace is now queryable. A one-click setup in Settings lets AI agents and CLI tools search your dictations, memos, meetings, and journal. A new CLI tool also lets you query from the terminal.

Meeting recording arrives with a dedicated Meetings tab in Settings. Onboarding now highlights key features, Smart Links are free for everyone, and the new branded DMG installer makes a better first impression.

Auto-paste now detects when macOS silently revokes accessibility permissions after an update, and shows clear instructions to re-grant access instead of failing silently.

Dictated emails now paste with proper line breaks in Gmail and Outlook, and text transformations no longer leak context from previous dictations.

Transform selected text now works with any spoken instruction, not just specific keywords. This release also fixes subscription sync issues that could block Pro features and improves streak and feed reliability.

Dictation no longer suppressed in apps where the native accessibility check incorrectly reported no text input — now falls back to the legacy check for any ambiguous result.

Selection capture now reads selected text directly via the Accessibility API instead of touching the clipboard. Long recordings with leading or trailing silence now retry on the trimmed speech window, and the dictation pipeline runs 50–200ms faster.

Fixes for dictation output being contaminated by VSCode identifiers, the trigger key getting overwritten on fresh install, and a false positive in the launch integrity check.

Notch hover-to-open now works reliably — replaced the SwiftUI .onHover handler (broken at the system window level) with a global mouse monitor that fires unconditionally, using the exact notch hit area instead of the full window frame.

The biggest update since launch. Resonant's entire interface has moved into your Mac's notch — a native overlay that lives at the top of your screen and gets out of the way completely when you're not using it.

Every interaction now happens in that strip of hardware: the waveform while you speak, the processing pulse as your words are formatted, the result landing with a snap. Celebrations for milestones. Audio cues for every state. It's the same dictation engine — it just feels like part of the OS now.

A cleaner layout across settings and the feed, with search, filtering, and keyboard navigation. Agent controls are now gated until an LLM is connected so the interface stays clean for everyone else.

A cleaner settings experience, a smoother onboarding flow, and a fix for the overlay stealing focus from your active app during dictation.

Resonant now handles long dictations up to 5 minutes. Audio is decoded in the background while you speak, so the wait after you stop is the same whether you dictated for 10 seconds or 5 minutes.

Resonant is now completely free — a private alternative to big tech dictation with zero data retention. Your voice is processed entirely on your Mac. Nothing is recorded, stored, or sent anywhere. Not to us, not to anyone.

This release is also noticeably snappier. A rewritten text engine, smarter memory management, and faster startup across the board make dictation feel instant.

Reliability improvements to keep Resonant rock-solid, even when you start dictating the instant the app launches.

Make Resonant yours. Name your assistant, choose your transcription style during setup, and enjoy faster repeated phrases thanks to smart tone caching.

A redesigned onboarding experience, built-in support, and under-the-hood reliability improvements — plus the start of our beta rollout.

The first release of Resonant. A private voice dictation app that runs entirely on your Mac — no cloud, no data leaving your machine.

Resonant processes your voice locally using on-device models, then types the result directly into whatever app you're using. It works across your entire system — notes, code editors, browsers, messages.