Drop a transcript. Get action items, decisions, and follow-ups.

In production, meetings are ingested automatically via webhook from Zoom, Teams, or Google Meet the moment they end. A speaker attribution layer maps each utterance to a named participant. Before the AI sees anything, a deterministic pre-processor scans the transcript for ownership events: outbound assignments, peer intercepts, self-commits, and cancellations. It resolves conflicts in code — if a task is reassigned mid-meeting or explicitly dropped, the earlier assignment is removed before extraction begins. The AI then extracts decisions, action items with dependency-aware due dates, follow-ups, and sentiment. Results push to CRM, calendar, and task systems via OAuth without any manual steps.

Paste any meeting transcript to see the full extraction pipeline. The left panel attributes each line to its speaker with color-coded pills. The right panel shows the structured brief: summary, key topics, decisions, action items with dependency notes where applicable, follow-ups (only unresolved ones), and sentiment. Action items reflect final ownership — reassignments and cancellations from the conversation are already resolved.