Afterglow — Stay in the moment

A call ends. Something else begins. A booking to enter, a confirmation to send, an allergy to remember, a briefing for the next call. The current AI answer is to replace the human with a voice agent — but the reason a customer dials a small business is the human. Replace the human, kill the moat. We don't replace the call. We replace the after.

Afterglow is a drop-in phone app for booking-driven SMBs — restaurants, dental clinics, body shops, hair salons, dog groomers, garages, tutoring studios. The operator picks up live. The moment they hang up, an agentic loop runs in the background and lands a structured extraction + a one-sentence next-call briefing on the customer's record. The next time the same number calls, the operator's screen already knows the context — gluten-free, anniversary, Bichon Frisé, previous diagnosis — without anyone typing.

Under the hood: Speechmatics transcribes with diarization on; a Gemini 3 Flash agent in Google ADK runs up to 12 turns over the transcript, calling typed tools — re-read a segment, query the customer's prior-call memory through Vultr's Vector Store via Serverless Inference RAG, execute booking / WhatsApp / calendar / payment / CRM / review action tools whose payloads are validated against a per-template JSON Schema, flag for review, or finalize. Each action result feeds back into the next turn so the model can self-correct, capped at two attempts. The full reasoning trail is a first-class UI surface, auditable turn by turn.

Single-tenant. Deployed on Vultr (Cloud Compute + Managed Postgres + Vector Store + Serverless Inference) through Coolify, auto-deploy on push to main. MIT.

USP vs CallRail / Aircall / Dialpad AI: we don't replace the human. We replace the after. Stay in the moment. We handle the after.

Note: the demo URL is substituted with the GitHub repository link.