Skip to main content
Steno is a macOS app that records, transcribes, and summarizes meetings using AI models that run entirely on your device. By default there are no cloud uploads and no accounts required; the only outbound requests are model downloads during setup, periodic update checks, and anonymous usage analytics (no meeting content) that you can turn off in Settings. Your audio, transcripts, and summaries are stored locally in ~/Library/Application Support/stenoai/ (you can change the location in Settings).

​
How it works

Steno runs a local pipeline on every recording:
  1. Record — always captures microphone audio, with system audio (both sides of a call) as an optional add-on toggle
  2. Transcribe — runs a local speech-to-text engine to produce a verbatim transcript with speaker labels when system audio is on. Steno ships two engines: Parakeet (the default, which also shows a live transcript while you record) and Whisper (an optional alternate you can switch to in Settings)
  3. Summarize — passes the transcript to a small language model running via Ollama on your Mac to produce structured notes: summary, key topics, and action items
  4. Query — ask natural-language questions against any saved note from the inline ask bar
Recording, transcription, and summarization all run offline. Steno does not send meeting content to any external server unless you explicitly configure a non-local backend (a cloud model such as OpenAI, Anthropic, AWS Bedrock, or a custom OpenAI-compatible endpoint, or a remote Ollama / organization adapter) as an optional alternative to the local model.

​
Who it’s for

Steno is built for professionals who work with confidential audio:
  • Healthcare — clinical consultations, case reviews, MDT meetings
  • Legal — client calls, depositions, strategy sessions
  • Finance — investment committee meetings, client briefings
  • Any role where recordings leaving your device is not acceptable
It is also used by developers, researchers, and individuals who simply prefer their data to stay on their own hardware.

​
What Steno is not

Steno is not a meeting bot. It does not join your video calls as a participant, does not require calendar access, and does not require you to share a link with other attendees. It records audio from your Mac’s audio system — no bot appears and the app does not notify other participants. Consent obligations remain yours to manage under your local laws and policies.

​
Supported platforms

macOS 14.4 (Sonoma) or later, on Apple Silicon (M1 and newer). Intel Macs are not supported since v0.4.0.
No, in the default local configuration. After the initial download and model installation, Steno runs offline — transcription and summarization happen on your Mac. The app still performs periodic update checks and sends anonymous usage analytics (no meeting content, can be turned off in Settings), and if you opt into a cloud summarization model (OpenAI, Anthropic, AWS Bedrock, or a custom OpenAI-compatible endpoint) it will reach that provider’s API for summary generation only.
All data is stored in ~/Library/Application Support/stenoai/. This includes recordings (recordings/), transcripts (transcripts/), and summaries (output/). Nothing is stored remotely.
Yes. The source code is available at github.com/ruzin/stenoai under the MIT license.
An Apple Silicon Mac (M1 or later) with at least 8GB of RAM — the default models total roughly 7GB on disk. Larger optional summarization models need more RAM and disk space.