Changelog
What's new in Whisperstream.
Every release, grouped and dated. The running record of what shipped and what changed.
- v2.0.3
Cleaner long recordings and a more reliable start
- Dictation: Fixed repeated words in long recordings.
- Local enhancement: Stays ready during very long recordings, so they still get cleaned up.
- Dictation: Fixed a problem when you start recording right as the app opens.
- Mac only:Dictation: Fixed a hang when stopping a recording.
- Windows only:Startup: Fixed a signing issue that could stop the app from opening.
- v2.0.2
Recording reliability and clearer bug reports
- Bug reports: Report a bug now prompts you to save logs before opening GitHub.
- Mac only:Dictation: Fixed an intermittent hang when stopping a recording.
- Windows only:Dictation: Improved recovery after recording errors.
- v2.0.1
Lower memory voice and native window controls on Mac, plus fixes
- Agent Speak: Fixed duplicate Claude Code replies and an old reply playing when the app starts.
- Mac only:Lower memory voice: Agent Speak can now use the Lower memory voice (Supertonic 3).
- Mac only:Native window controls: Standard traffic-light buttons for closing and minimizing.
- Mac only:Agent Speak: Fixed a click at the end of spoken replies.
- v2.0
Whisperstream 2.0: free, open source, and now on Mac
- Every feature is included, with no trial, license key, or activation. The source is on GitHub.
- Whisperstream now runs on Apple Silicon Macs as a beta, with the same features as Windows.
- Stream-1 Mini, our own on-device cleanup model, is the new default for Local enhancement, with Chat and Email styles. It's about a 460 MB download and runs on computers with 4 GB of memory or more.
- Dictionary Boost helps names and specialist terms come out spelled right, even when they sound like other words. Import a whole list from a CSV file. Requires AI enhancement.
- Cohere Transcribe is a slower speech model for when accuracy matters most, especially in European languages. Choose it in manual model selection.
- A recovery key, plus an optional recovery password, gets your encrypted transcript history back on a new computer or after a reset.
- Connect Claude Code or Codex to Agent Speak in one click, pause it anytime, choose spoken summaries or full responses, and type or dictate replies from recent conversations.
- Recordings with no speech are skipped, repeated phrases are kept, Whisperstream reconnects reliably after sleep, and modifier keys no longer stick after push-to-talk. It also uses less memory in the background and shuts down cleanly with Windows.
- Pin Mode, experimental since 1.3, has been removed.
- v1.6
Free trial extension.
- v1.5.1
Agent Speak summaries and refreshed cloud models
- See concise written agent replies in the Agent Speak overlay.
- Choose from a refreshed, curated selection of cloud models for AI cleanup.
- Improved recording startup and Agent Speak playback reliability.
- Fixed a recording indicator glitch after Agent Speak.
- v1.5
Agent Speak for Codex, clearer enhancement feedback, and more control
- Agent Speak now supports Codex, with guided onboarding for Codex and Claude Code. The Whisperstream CLI can wrap sessions so you can reply anytime.
- See which enhancement profile is active while recording and processing, with clear timeout and error feedback. Transcript History now separates Original and Enhanced text and records the profile used or why enhancement fell back.
- Administrators can enforce Lockdown across configuration changes, restarts, updates, and reinstalls. License activation now recognizes administrator-installed Windows certificate roots on managed networks.
- Choose between NVIDIA Parakeet and Qwen3 ASR when both support your selected language.
- v1.4
Click to record from your desktop
- Turn on Keep control visible in Controls to leave a compact microphone control on your desktop between recordings. Click it to start, then cancel or finish from the recording overlay.
- Press Escape during active dictation to discard it before Whisperstream transcribes or types anything.
- Modifier-based shortcuts now respond only to the configured combination, with no extra modifiers pressed.
- Improved local data cleanup for temporary files, diagnostics, cached app and website details, and expired encrypted transcript history.
- A new documentation center brings setup instructions, feature guides, and privacy information together in one place.
- The recording overlay is more responsive, with additional stability fixes across the app.
- v1.3
Live Preview, Pin Mode, and a voice for Claude Code
- Live Preview shows your words in the recording overlay as you speak, so you can tell dictation is working before you stop. Switch it on in the Controls tab.
- A new Experimental tab holds two features that are still finding their footing. Pin Mode pins one window as the destination for your dictation, so text lands there even while you work in another window. If the pinned window goes away, nothing is sent, rather than landing somewhere you did not intend.
- Agent Speak, also experimental, is a companion for Claude Code. It reads replies out loud and lets you answer by voice with your existing dictation shortcut.
- AI cleanup now preserves line breaks in enhanced text.
- You can now override the AI cleanup timeout in Advanced settings.
- v1.2
Dark Mode, Email Mode, and Lockdown Mode
- Dark Mode adds a full dark theme alongside the light one. Pick light or dark in the Home tab.
- Email Mode is a new AI enhancement mode that formats your dictation into a ready-to-send email, with a greeting, body, and sign-off. It can switch on by itself in your email app and add your saved signature.
- Lockdown Mode adds extra protection for privacy-sensitive use cases. Your transcript history is password-protected and encrypted, AI enhancement stays fully on-device, and diagnostics are off. Designed for HIPAA-conscious and confidentiality-sensitive work.
- v1.1
Additional dictation languages
- Added support for 14 more dictation languages: Arabic, Cantonese, Chinese, Filipino, Hindi, Indonesian, Japanese, Korean, Macedonian, Malay, Persian, Thai, Turkish, and Vietnamese.
- AI cleanup and custom prompts now work with any OpenAI-compatible provider.
- Choose where speech models are downloaded.
- Start on login now works reliably.
- v1.0.1
Type into remote desktops and screen readers
- Keystroke input mode sends text character by character instead of pasting it. This can work in Citrix, RDP, and screen-reader workflows when endpoint, administrator, security, privilege, session, and destination policies permit simulated input.
- A new shortcut pastes your most recent transcription before AI cleanup, so the original is one keypress away.
- v1.0
Per-app cleanup and a private transcript history
- Automatic profile switching gives each app or website its own AI cleanup style. Whisperstream detects the active window and applies the right one.
- Transcript history keeps a searchable, private record of everything you dictate, with audio playback. It stays encrypted on your device.
- File import lets you drop in an existing audio file to transcribe it.
- Set separate shortcuts for push-to-talk and toggle recording.
- The free trial is now unlimited dictation for 7 days.
- v0.9
AI cleanup, now fully on-device
- AI Enhancement can now run through an integrated, downloadable local model on PCs with a capable graphics card. No API key or cloud cleanup request is needed for that mode.
- v0.8
AI cleanup, bring your own key
- Optional AI Enhancement sends transcriptions to Gemini, OpenAI, Anthropic, or a local Ollama model for cleanup before they're typed. It is off by default, and you bring your own API key.
- Pick Chat or Code mode, or write a custom style prompt.
- v0.7
Automatic updates and flexible shortcuts
- Automatic daily update checks, plus a Check for updates button on the About page.
- Auto-update now closes the app cleanly before installing.
- Your push-to-talk shortcut now accepts extra mouse buttons and modifier keys.
- Clearer errors when recording can't start, plus stability fixes.
- v0.6
Faster dictation
- Parakeet speech model now stays loaded between recordings, so dictation starts instantly.
- Faster app startup and close.
- v0.5
Signed installer and in-app Pro
- Whisperstream now ships with a verified Microsoft publisher signature, so Windows no longer warns about an unknown publisher.
- Buy and activate Pro without leaving the app.
- Click-through end user license agreement on first launch.
- Release notes now appear in the in-app update popup.
- v0.1
First release
- Push-to-talk dictation powered by NVIDIA Parakeet, running locally on your CPU.
- Live HUD waveform while you record.
- Custom dictionary for word and capitalization overrides.
- Usage dashboard with transcription, word, and recording-time stats.
- Auto-update via signed installer, with one-click log export for bug reports.