Capture a session

A session is the unit of capture in SmartNotes. You start it, speak naturally while you work, and stop it when you are done. Every spoken thought becomes a note with a screenshot, transcript, timestamp, and an issue tag.

Start a session

1. Go to the Capture tab.
2. Confirm the screenshot source, microphone, and output folder.
3. Click Start session.

SmartNotes is now listening. It does not record continuously — it waits for speech.

How capture works

VAD onset (speech energy above threshold for 300ms)
  → one screenshot is taken
  → audio starts buffering

Speech continues (with pauses)
  → audio keeps appending to the same note

VAD offset (silence for 700ms)
  → audio blob is finalized
  → ffmpeg converts WebM/Opus to WAV 16kHz mono
  → whisper-cli transcribes the audio
  → IssueClassifier tags the note
  → note is saved and pushed to the UI

Cooldown 7 seconds:
  if new speech starts within this window, it continues the same note
  without creating a new screenshot.

The 7-second debounce is what makes one thought stay as one note, even if you pause between sentences.

Note tags

Each note gets a quick local heuristic tag:

These tags are local helpers. When notes are sent to Cradle, the triage agent there makes the final classification.

Review notes

Open the Notes tab to see sessions and individual notes. You can:

• Filter notes by tag.
• Edit a transcript or reclassify a note.
• View the screenshot that was captured with each note.
• Select notes and run Summarize to get a Markdown summary plus issue cards.

Summarization

Summarization can run locally through Cradle if connected, or optionally through OpenAI (gpt-4o). It takes the captured notes, cleans them, groups related items, and produces:

• A short Markdown summary of the session.
• Issue cards that can be submitted to Cradle for triage.

Summaries do not auto-send anything — you review the issue cards first.