Skip to main content

​
About Steno

Yes. Steno is free and open source under the MIT license, which places no restriction on who can use it or at what scale — including enterprise and commercial use. The source is available on GitHub, so you can read exactly what it does, and there is no account to create. On top of the free app, a paid managed Enterprise deployment tier (admin controls, organization-specific features) is a separate, optional offering — details coming soon.
Yes. After the first-run model download, Steno works fully offline — transcription and summarization run on-device, so no meeting audio, transcript, or summary is ever sent to a server. The app periodically checks for updates and sends anonymous usage analytics (no meeting content) by default; both can run without affecting offline recording, transcription, and summarization.
Steno runs on macOS (Apple Silicon, M1 and later — Intel Macs are not supported since v0.4.0) and on Windows 10/11 (x64) in alpha, with the full on-device pipeline verified working. There is no iPhone or web version.
Steno saves your notes as plain Markdown files under ~/Library/Application Support/stenoai/output/ (macOS) or %APPDATA%\stenoai\output\ (Windows). Because they are ordinary files, you can move them into Notion, Obsidian, or any other tool by hand or through your own automation.
Everything Steno stores lives under ~/Library/Application Support/stenoai/ (macOS) or %APPDATA%\stenoai\ (Windows) — recordings, transcripts, and summaries. To back up your data, copy that folder. To delete it, remove that folder. Nothing is stored remotely, so there is no cloud copy to clear.

​
Privacy and data

No. Steno processes all recordings using AI models that run locally on your Mac. No audio, transcript, or summary is transmitted to any external server unless you configure an off-device summarization backend (cloud model, private server, or organization adapter). The app also sends anonymous usage analytics by default — see the telemetry question below.
All data is stored locally:macOS: ~/Library/Application Support/stenoai/Windows: %APPDATA%\stenoai\ (e.g. C:\Users\<your-name>\AppData\Roaming\stenoai\)
  • recordings/ — raw .webm audio files
  • transcripts/ — verbatim transcript text
  • output/ — Markdown summaries and notes
Nothing is stored remotely.
Yes, anonymous usage analytics are on by default — event names (e.g. “recording started,” “summarization completed”), error types, and an anonymous install ID. This never includes audio, transcripts, notes, or summaries. Turn it off in Settings → Advanced.
Steno is designed for professionals handling confidential audio. Because processing is entirely local, it is suitable for use cases where sending audio to a cloud service is not acceptable — clinical consultations, legal calls, financial meetings. Review your organization’s specific policies before using any recording tool with regulated data. See Confidential use cases for more detail.

​
Installation and setup

Steno is signed but distributed outside the Mac App Store. On first launch, macOS Gatekeeper may block it. Go to System Settings → Privacy & Security and click Open Anyway next to Steno. This is a one-time step.
Steno’s Windows alpha builds are not digitally signed yet. Windows SmartScreen flags unsigned executables with a warning. To proceed, click More info and then Run anyway. The app is safe to run — we will sign the installer before the general availability release.
The default setup uses roughly 7GB: the Parakeet transcription model (~572MB) plus the default Gemma summarization model. Plan for around 8GB free during install. Larger summarization models take more space. Recordings accumulate over time in ~/Library/Application Support/stenoai/recordings/ (macOS) or %APPDATA%\stenoai\recordings\ (Windows).
Steno updates automatically in the background. When an update is ready, you will see a notice in the app. The update installs on the next quit — you do not need to download a new DMG.
Delete Steno from your Applications folder (macOS) or uninstall via Settings → Apps (Windows). To also remove all recordings, transcripts, and model files, delete ~/Library/Application Support/stenoai/ (macOS) or %APPDATA%\stenoai\ (Windows), and ~/.ollama/ (if you want to remove the Ollama models too).

​
Recording

Yes. Steno captures your Mac’s system audio and your microphone, so it is independent of the calling app — Zoom, Microsoft Teams, Google Meet, or anything else playing audio on your Mac. In-person meetings work too, captured through the microphone.
No. Steno records audio locally from your Mac’s audio system and never joins the call as a participant. No bot appears, and no other attendee is notified by Steno. Recording-consent obligations remain your responsibility.
Steno can capture system audio — the audio playing through your Mac’s speakers or headphones, which includes the other participants on a call. See Recording for setup instructions. System audio capture requires a one-time permission setup.
Yes. Steno records from your Mac’s microphone, which will pick up voices in the room. For best results, place your laptop centrally. In-person recordings are mic-only, so they have no [You] / [Others] speaker labels.
No. Steno captures audio from your Mac’s audio system without joining the call as a participant. Other attendees will not see a bot or receive a notification from Steno. Recording consent obligations are your responsibility.

​
Transcription and summaries

Steno defaults to the Parakeet engine, which is fast on Apple Silicon and shows a live transcript while you record. If you have switched to the Whisper engine in Settings → Transcribe, switching back to Parakeet is usually faster; Whisper transcribes only after you stop recording.
Switch to a larger summarization model. In Settings → AI, try qwen3.5:9b, gemma4:12b-it-qat, or gpt-oss:20b. Larger models produce more detailed and accurate notes. You can regenerate notes for any existing recording from its detail view.
Yes. The default Parakeet engine transcribes 25 European languages, including English, French, German, Spanish, Italian, Dutch, Portuguese, Russian, and Polish. A curated subset of six (English, French, German, Spanish, Dutch, Portuguese) is available to pin as the summary output language. For Japanese, Chinese, Korean, Hindi, Arabic, and the full set of 99 languages, switch to the Whisper engine in Settings → Transcribe, which auto-detects the language from the audio.

​
Models

Transcription: two on-device engines — Parakeet (the default) and Whisper (optional, its single large-v3-turbo model), selected in Settings → Transcribe. Summarization: a local model via Ollama, with gemma4:e2b-it-qat as the default. See Transcription models and Summarization models for the full comparison.
Yes. Steno optionally supports OpenAI, Anthropic, AWS Bedrock, or a custom OpenAI-compatible endpoint as an alternative to the local model. Configure this in Settings → AI, under Cloud API. It is off by default. Note that this sends your transcript and any typed notes (not audio) to the selected provider’s API.