Core dictation is free, runs on-device, and has no word limits. A license adds everything else, such as AI features, media transcription, and automations.
Free, always
- Local transcription. Turn your wifi off, it still works. ANE supported on Apple Silicon.
- Custom dictionary. Teach it names, brands, or terms.
- Auto Dictionary. It picks up your custom words on its own.
- Replacements. Say "my address," get 221B Baker Street.
- History and search. Find anything you've dictated.
With a license
Free during the 14-day trial, then any license:
- Library. Drop in audio or video, scrub the synced transcript, assign speakers, export to
.txt,.md,.srt, or.vtt. - AI Cleanup. Polish dictated text with your own LLM.
- Edit Mode. Highlight text, say what you want, and watch it rewrite in place.
- Personalization. Different tones per app or site, with snippets for dynamic context.
- Local API. An OpenAI-compatible speech endpoint, running on your machine.
- CLI. An optional
glimpsecommand for the terminal.
Configure AI writing in Settings → Providers. Speech models live in Settings → Models.
- Raycast (coming soon). Search dictations, transcribe files, switch models, and more, without leaving Raycast. Requires a license.
Want to build your own? See the CLI guide.
| Edition | Price | For |
|---|---|---|
| Personal | $24.99 one-time | You, on up to 5 personal devices |
| Commercial | $48 / seat / year | Work use, one seat per person, one device |
Start with the 14-day trial, then buy or paste a license key in Settings → Account.
Transcription stays on-device by default.
Glimpse sends anonymous usage telemetry via PostHog EU to help prioritize development. It's tied to a random install ID, not your identity, and stored in the EU.
- Collected: app version and platform, launches and uptime, durations and counts, country, and bounded error/crash categories. A crash also records a code location (source file and line, or module and offset) so we can find the bug.
- Never sent: transcripts, audio, API keys, prompts, raw error text or stacks, file paths or names, microphone names, provider endpoints, your IP, or anything personally identifiable.
Opt out anytime in Settings → App; opting out sends one final ping, then nothing, ever.
Enabling an external speech or LLM provider sends audio or text directly to that provider. Your API keys stay local.
For the full picture, see the analytics wiki or analytics.rs.
The source code is licensed under AGPL-3.0. If you distribute Glimpse or run it as a network service, you must make your modified source available under AGPL-3.0.
Trademarks. The Glimpse name and logo are not part of the AGPL-3.0 license. Forks and redistributions must use a different name and icon.
Want to help? The Contributing Guide covers everything from translations to code to bug reports.
Questions, bugs, or feedback: hello@tryglimpse.cc or GitHub Issues.
Lokalise (localization platform, OSS supporter)- Tauri (app framework)
- Glimpse-Speech (MIT, local transcription engine)
- whisper-rs (Unlicense, Rust bindings for Whisper)
- parakeet-rs (MIT OR Apache-2.0, ONNX Runtime bindings for Parakeet)
Speech models are downloaded in-app from Hugging Face; the live list lives in Settings → Models. By family:
- Whisper GGML (MIT), via
ggerganov/whisper.cpp - Distil-Whisper GGML (MIT, English-only), via Pomni's conversions of
distil-whisper - Parakeet TDT ONNX (CC-BY-4.0), via
istupakov - Parakeet Unified ONNX (CC-BY-4.0, English-only), via
bobNight - Nemotron Streaming ONNX (NVIDIA Open Model License), via
altunenes/parakeet-rs