Compatibility & Requirements

  • Platforms: Desktop v1.10.0+ (macOS/Windows). Linux support pending.
  • Hardware: GPU recommended for advanced features (e.g., Full Diarization).

Transcription & Media Sources

  • Sources: Live recording (Zoom, Meet, Teams, local microphone) or file upload.
  • Real-Time Transcription: Disabled by default. Toggle in settings (requires high-end hardware).
  • Interactive Transcript:
    • Double-click text to edit transcription.
    • Click transcript segments or speaker names to jump to that timestamp in the audio player.

Speaker Diarization (Identification)

Configure under Preferences (Sliders icon, top-right):

Diarization ModePerformance CostDescription
No DiarizationNoneRaw transcript without speaker labels.
Basic DiarizationMinimal (Default)Differentiates β€œUser” (you) from β€œOther” (all other audio combined).
Full DiarizationHigh (+20-40% time)Identifies individual speakers by unique voice characteristics.

Management: Click a speaker label to rename across the entire transcript.


Meeting Summaries

Summaries use the active system model configured in Settings > AI Providers > LLM.

Templates

Access via the 3-dot menu next to the recording player.

  • Pre-built Templates: General Meeting, Sales Call, Engineering Meeting.
  • Custom Templates: Create per-meeting or load existing. Click Apply Summary Template to send to the LLM.
  • Regenerate Summary: Select Regenerate Summary from the 3-dot menu to update after editing transcripts or templates.

Action Items & Agentic Follow-ups

Proposes tasks based on summary/transcript using active Agent Skills (MCPs, Agent Flows).

  • Execution: Select Run to execute. No actions run automatically.
  • Inspection: Click View Arguments to inspect parameters before execution.
  • Regenerate: Select Regenerate Action Items from the 3-dot menu to update.

Ask Questions Tab

  • Opens a dedicated workspace containing the embedded transcript for interactive QA.
  • Accessed from the right sidebar.
  • Searches transcripts and summaries semantically using a local, on-device vector database and embedding model.