Casper is an AI meeting assistant that handles meetings end to end and, uniquely, measures team dynamics. From chat you can schedule meetings (it creates the Google Calendar event, Meet link, and recording bot in one shot), auto-record an entire calendar, and send or control bots on any Zoom/Meet/Teams call. When a transcript lands, a webhook automatically generates summary, decisions, action items, topics, and soundbites, then notifies you. Its differentiator is reading how a team works, not just what was said. A meeting-health dashboard shows talk-time per person, interruptions, silences, monologues, and a participation balance score, all deterministic timestamp math with zero LLM tokens. A single Fireworks insight call turns those metrics into a manager-facing read. On-demand acoustic tension detection decodes audio in the browser (WebCodecs, no upload) to tell real tension from casual agreement, and a behavioral layer interprets what each tense moment reveals. Trends over time reveal who is fading out or taking over the room. Screen Intelligence adds a vision layer: a deterministic pre-filter scans the whole recording and picks only the handful of frames that matter, hard-capped at 12, so a 90-minute call becomes about 8 vision calls instead of thousands of frames. Built on Next.js 16, React 19, Mastra, and Fireworks AI running on AMD hardware, with Postgres RLS multi-tenancy, Svix-signed webhooks, and self-healing recovery crons.
Category tags: