FollowThrough

Meetings generate decisions and action items that quietly die in nobody's inbox. FollowThrough is a three-stage agent pipeline — not a single LLM call — that fixes this: an **Extract** agent pulls candidate action items out of a raw transcript; a second, independently-prompted **Verify** agent acts as a skeptic and cross-checks every item against the source text, flagging or dropping anything not actually grounded in what was said; a **Draft** agent then writes a ready-to-send follow-up email and Slack message for everything that survives verification. The UI shows all three stages live, so you watch the agent extract, doubt itself, and only then act — instead of trusting a black box. It runs on Fireworks AI, is fully containerized, and generalizes past meetings to any workflow where LLM output needs to be trustworthy enough to actually send: support tickets, sales calls, standups, incident postmortems.