llms.txt
Compare with other sites' llms.txt files →# Whisperstream
Whisperstream is private local dictation software for Windows, with a beta for Macs with Apple silicon. It captures your voice with a push-to-talk hotkey, transcribes it entirely on your computer using an on-device speech model, and inserts the result into the app you are using. Your microphone audio is not sent to a transcription service. Optional AI enhancement is off by default and can use a downloadable Local model, Ollama, or a provider the user configures with their own key. Free and open source, with no subscription or account.
## Key facts
- Platform: Windows 10, Windows 11 (64-bit); Mac beta for Apple silicon on macOS 14 Sonoma or later (Intel Macs are not supported)
- Price: Free, with every feature included and no subscription or account
- Source code: [github.com/lanreal-io/whisperstream](https://github.com/lanreal-io/whisperstream)
- Publisher: Lanreal Technologies Inc., Toronto, Canada
- Product page: [whisperstream.io](https://whisperstream.io)
- Bug reports and feature ideas: [GitHub Issues](https://github.com/lanreal-io/whisperstream/issues/new/choose)
- Private support: [support@whisperstream.io](mailto:support@whisperstream.io)
- Publisher contact: [contact@lanreal.io](mailto:contact@lanreal.io)
- GitHub org: [github.com/lanreal-io](https://github.com/lanreal-io)
- Company website: [lanreal.io](https://www.lanreal.io/)
## Features
Full feature page: [whisperstream.io/features](https://whisperstream.io/features)
- On-device transcription: Speech recognition happens locally using the model selected for the dictation language. Automatic selection uses NVIDIA Parakeet or Qwen3 ASR. Cohere Transcribe is an optional Manual choice for supported languages. Microphone audio is not uploaded for transcription.
- Lockdown Mode: Keeps Whisperstream's transcription, AI cleanup, and history on your computer, limits cleanup to Local, and can password-protect saved history. Safe local diagnostics remain available for user-controlled bug reports and are never uploaded automatically. It does not control the app receiving inserted text. Agent Speak processing is suspended while Lockdown is effective. On Windows, organizations can enforce it as Managed Lockdown.
- Push-to-talk hotkey: Hold a hotkey to dictate, release to stop. Default is Right Shift, remappable to any key; a toggle mode is also available. No admin rights needed. On a Mac, shortcuts need Accessibility access.
- Text insertion: Uses clipboard paste or simulated keystrokes to deliver text to compatible fields. Some applications and managed or remote environments can block delivery.
- Transcript history: Encrypted text and audio history is on by default on Windows and off by default on Mac, with seven-day retention. A recovery key restores saved history if the device key is lost. Users can turn text or audio saving off, change retention, search, replay, copy, save an audio copy, and delete records locally. History is not cloud-synced.
- Audio file import: Transcribes audio files you already have through the selected on-device speech model. Whisperstream does not upload the source audio. If online enhancement is enabled, it can send the resulting transcript to your configured provider.
- Optional AI cleanup: Off by default. Removes filler words, fixes grammar and punctuation, and applies spoken self-corrections. The Local provider runs Stream-1 Mini (the default, a fine-tuned Qwen3 0.6B model for English Chat and Email cleanup) or the Customizable Gemma 4 E2B model on the user's computer. Ollama, Gemini, OpenAI, Anthropic, and configured OpenAI-compatible providers are also supported.
- Automatic profile switching: Gives each app or website its own enhancement profile and applies it automatically by detecting the focused window. Start from an included profile or write your own; route by app or by a specific site, with a default everywhere else.
- Custom dictionary: Boost gives AI enhancement spelling hints for names and technical terms; filters remove unwanted words or phrases; replacements substitute recurring recognized terms. Filter and replacement matching is case-insensitive and applies to completed dictation. Word lists can be imported from a CSV file.
- Agent Speak for coding agents: Connects to Claude Code and Codex on the user's computer. When an agent finishes, Whisperstream can show a short written summary, read it aloud with an on-device voice, or open the full response, and the user can reply by dictation. One-click connect, can be paused, and off until set up. Connected agents keep using their own model services.
- 39 languages: Transcribes 39 selectable languages on-device across English-only and multilingual speech-model families. A recording should contain one language, and changing to a language unsupported by the active model can require another model download.
- Offline core: After the required speech model is downloaded, recording and speech recognition work without a connection. The downloaded Local enhancement model can also work offline after setup. Model downloads, configured online enhancement providers, and connected coding agents still use the network. The official installer also checks for signed updates, installed only after the user accepts; Lockdown Mode stops these checks, and builds from source make no such request. Whisperstream makes no license-validation or telemetry requests.
- Authenticode-signed Windows installer: Windows can verify Lanreal Technologies Inc. as the publisher, and IT teams can use that exact name in managed environments. SmartScreen may still show a reputation warning.
- Custom style prompt: Give AI cleanup your own prompt with custom formatting or style instructions (tone, terseness, spelling, bullet formatting, house style). Custom prompts run with the Customizable local model or a configured provider; Stream-1 Mini uses its built-in prompts.
- System-volume ducking: Optionally lowers (to 20 percent) or mutes system volume while you dictate, then restores it on release.
- Runs on CPU, no GPU required: Core transcription runs on a normal processor; a GPU is only used for the optional AI cleanup.
## Canonical pages
- [Home](https://whisperstream.io/)
- [Features](https://whisperstream.io/features)
- [About](https://whisperstream.io/about)
- [End User License Agreement (EULA)](https://whisperstream.io/eula)
- [Privacy Policy](https://whisperstream.io/privacy)
- [Website Privacy Policy](https://whisperstream.io/privacy-website)
- [Subprocessors](https://whisperstream.io/subprocessors)
- [Terms of Service](https://whisperstream.io/terms)
## Product documentation
- [Install and configure Whisperstream](https://whisperstream.io/docs)
- [App basics, updates, and uninstall](https://whisperstream.io/docs/getting-started/app-basics)
- [Recording controls and last-transcript actions](https://whisperstream.io/docs/dictation/recording-controls)
- [Languages and speech recognition models](https://whisperstream.io/docs/dictation/languages-models)
- [Microphone, playback, and system audio](https://whisperstream.io/docs/dictation/audio)
- [Live Preview](https://whisperstream.io/docs/dictation/live-preview)
- [Text insertion and compatibility](https://whisperstream.io/docs/dictation/text-insertion)
- [Dictionary Boost, filters, and replacements](https://whisperstream.io/docs/dictation/dictionary)
- [Local, Ollama, and cloud enhancement providers](https://whisperstream.io/docs/enhancement/providers)
- [Profiles, Email mode, app detection, and site detection](https://whisperstream.io/docs/enhancement/profiles-routing)
- [Transcript and audio history](https://whisperstream.io/docs/data/transcript-history)
- [Lockdown Mode](https://whisperstream.io/docs/data/lockdown-mode)
- [Managed Lockdown for organizations](https://whisperstream.io/docs/data/managed-lockdown-deployment)
- [Agent Speak for Codex and Claude Code](https://whisperstream.io/docs/workflows/agent-speak)
- [Managed Codex and Claude Code sessions](https://whisperstream.io/docs/workflows/coding-agent-sessions)
- [Settings and data locations](https://whisperstream.io/docs/reference/settings-data-locations)
- [Install and recover models offline](https://whisperstream.io/docs/reference/model-recovery)
## Comparisons and alternatives
- [Whisperstream vs Wispr Flow](https://whisperstream.io/vs/wispr-flow): Side-by-side comparison of Whisperstream and Wispr Flow on price, privacy, and Windows performance, with a 5-step switching guide.
- [Whisperstream vs Windows Voice Typing (Win+H)](https://whisperstream.io/vs/windows-voice-typing): Side-by-side comparison of Whisperstream and the built-in Windows Voice Typing (Win+H) on price, privacy, reliability, and cloud-vs-on-device behavior.
- [Whisperstream vs Willow Voice](https://whisperstream.io/vs/willow-voice): Whisperstream vs Willow Voice on price (free vs $144/year), privacy, and AI rewriting; Whisperstream adds optional on-device AI cleanup.
- [Whisperstream vs Aqua Voice](https://whisperstream.io/vs/aqua-voice): Whisperstream vs Aqua Voice on price, privacy, and a controlled word-error-rate benchmark; literal on-device transcription with optional local AI cleanup vs Aqua's always-on Avalon rewriting.
- [Whisperstream vs OpenWhispr](https://whisperstream.io/vs/openwhispr): A focused, free and open-source Windows app with no cloud transcription path vs an MIT-licensed app offering three transcription modes.
- [Wispr Flow alternatives for Windows](https://whisperstream.io/alternatives/wispr-flow): Six Wispr Flow alternatives for Windows in 2026 (local, cloud, free, and paid), ranked by privacy and price.
- [MacWhisper for Windows](https://whisperstream.io/macwhisper-for-windows): MacWhisper is a Mac-only transcription app; Whisperstream is the Windows-native pick for on-device dictation and audio-file transcription (free and open source, NVIDIA Parakeet V3 on CPU). MacWhisper still goes further with speaker labels and subtitle and document export.
- [VoiceInk for Windows](https://whisperstream.io/voiceink-for-windows): VoiceInk is a Mac-only, open-source dictation app. Whisperstream is the free, open-source Windows-native option, with no device limit, on-device transcription and optional local or configured online enhancement.
- [Superwhisper alternatives for Windows](https://whisperstream.io/alternatives/superwhisper): Five Superwhisper alternatives for Windows in 2026, ranked for a Superwhisper user, with Whisperstream as the free, open-source local pick.
- [Dragon NaturallySpeaking alternatives for Windows](https://whisperstream.io/alternatives/dragon-naturallyspeaking): Five Dragon NaturallySpeaking alternatives for Windows in 2026 (local, cloud, free, and paid), ranked for a Dragon refugee.
- [Best dictation software for Windows](https://whisperstream.io/best-dictation-software-windows): The best dictation software for Windows in 2026, seven picks ranked by privacy and price.
- [Dictation software that works with screen readers on Windows](https://whisperstream.io/screen-reader-dictation-windows): Windows dictation apps tested with NVDA, JAWS, and Narrator, on two axes (does the hotkey fire, and does the text land). Several work; Superwhisper failed all three and Aqua Voice's default hotkey does not work with Narrator.
## Troubleshooting
- [Win+H not working: Microsoft-documented fixes](https://whisperstream.io/win-h-not-working): Seven Microsoft-documented fixes for Windows 11 Voice Typing (Win+H), in the order to try them, including the managed-org speech policy fix Microsoft's troubleshooter never mentions. Closes with the local-alternative pitch for Whisperstream.
## Guides
- [How to dictate in Word on Windows](https://whisperstream.io/dictate-in-word): Word's built-in Dictate button in the desktop app (Microsoft 365) and Word for the web (free with a Microsoft account), with voice commands, language settings, fixes, Windows' own dictation options, and dictating into Word with Whisperstream.
- [How to dictate in Google Docs on Windows](https://whisperstream.io/dictate-in-google-docs): Google Docs voice typing in Chrome or Edge on Windows: Tools > Voice typing, choosing the spoken language, microphone fixes, English-only voice commands, and dictating into Docs with Whisperstream.
- [Stream-1 Mini vs S1-mini](https://whisperstream.io/stream-1-mini-vs-s1-mini): A benchmark (November 2026) of two small, local dictation cleanup models, Whisperstream's Stream-1 Mini and Superwhisper's public S1-mini v076 checkpoint, on the same 1,930 development and regression tests. Reports pass, critical, major and minor rates overall and by kind of dictation, with three side-by-side cleanup examples.
- [Dictation app privacy: where your transcripts actually live](https://whisperstream.io/dictation-app-privacy): A six-app audit (July 12, 2026) of what dictation tools store and protect: transcript storage, audio at rest, encryption, history-off controls, locks, and cloud kill-switches. Open-source apps (Handy, VoiceInk, OpenWhispr) audited at source-code level with versions pinned; closed apps (Superwhisper, Wispr Flow) from their own published docs; Whisperstream's Lockdown Mode audited under the same rules. Headline finding: local is not the same as private; two local apps store plaintext transcripts and unencrypted audio of every dictation unconditionally, and cloud transcription is some local-capable apps' default.
- [HIPAA compliant dictation software: what actually matters](https://whisperstream.io/hipaa-dictation-software): A practical guide to using Lockdown Mode for sensitive dictation. It covers Whisperstream's local transcription, cleanup, and history scope, the separate destination-app and Agent Speak paths, and the user's remaining responsibilities.
- [Legal dictation software: where your client audio goes](https://whisperstream.io/legal-dictation-software): A guide to local dictation for confidentiality-sensitive work. It covers Whisperstream's local transcription and cleanup scope, the separate destination-app and Agent Speak paths, and the user's remaining responsibilities. Not legal advice.
- [Dictation for Citrix and RDP](https://whisperstream.io/dictation-for-citrix-and-rdp): Explains how to try Whisperstream's keystroke input when clipboard delivery is blocked. Delivery still depends on the remote session, administrator, security, privilege, and destination-app policies.