Skip to main content
This guide walks through making a short test recording so you can verify the full pipeline is working: record → transcribe → summarize → query.

Make a recording

1

Open Steno

Launch Steno from your Applications folder or from the menu bar icon.
2

Press Record

Click the Record button. Give the recording a name, or leave it blank — Steno will generate a title automatically from the transcript.Steno starts capturing audio immediately. You will see a live timer in the top bar.
3

Say something

Speak for 30-60 seconds. Introduce yourself, describe your work, or read a few sentences aloud — anything with clear speech works.
4

Stop the recording

Click Stop. Steno immediately begins transcribing.

Review the output

Transcription usually takes 10-30 seconds for a one-minute recording on Apple Silicon. When it finishes:
  • The transcript appears in the right pane with timestamped lines
  • A summary is generated below the transcript: a short paragraph, key topics, and any action items Steno identified
  • The recording appears in the sidebar under today’s date

Ask a question

Click the Ask bar at the bottom of any note. Type a question in plain English:
  • “What were the main points?”
  • “What actions did I mention?”
  • “Summarize this in one sentence.”
Steno passes your question and the saved note to the local model and streams the answer inline. No data leaves your Mac when using the default local model.

Where your files are saved

All output is stored locally:
File typeLocation
Audio~/Library/Application Support/stenoai/recordings/
Transcript~/Library/Application Support/stenoai/transcripts/
Summary / notes~/Library/Application Support/stenoai/output/
Summaries and transcripts are plain Markdown files. You can open, edit, and sync them with any tool you use for notes.
Transcription speed depends on the model and your hardware. On Apple Silicon, whisper-small processes roughly 5-10x real-time (a 10-minute recording takes 1-2 minutes). Switch to whisper-base in Settings for faster results at slightly lower accuracy.
Try a larger summarization model. In Settings → Models, switch from llama3.2:3b to qwen3.5:9b or deepseek-r1:14b. Larger models produce better-structured notes but take longer to run.
You need to enable system audio capture. See Recording → System audio for the one-time setup.