First Answer Agent: AI Front Desk

Customers hate waiting for a callback: voicemail, "someone will call you back", or an owner who is with another client or closed for the night and may not know the answer anyway. Many inquiries also arrive after hours. First Answer Agent is an AI front desk for appointment-based small businesses. It answers right on the page, any hour, and knows the business's hours, services, prices and openings. It qualifies each inquiry, checks the calendar, offers only open times, books it in the app (the customer gets a calendar file), schedules a follow-up, and brings in a person, and tells the customer so, for emergencies, angry customers and out-of-scope requests, with a phone alert to the owner. The AI can take only six actions, and code checks every one before it runs. Simple availability and pick requests are answered instantly by built-in rules; open-ended questions go to the AI model, and rules also answer if the model is slow or its reply is rejected. Safety checks apply to both. It runs on Vultr: Serverless Inference (glm-5.3-flash) for the AI, a Cloud Compute VM (vc2-1c-2gb, Atlanta) for the app, and Caddy for HTTPS. We tested it with 30 live inquiries on Nov 4, 2026 (a small sample of our own test messages). All 30 were answered, 6 booked by picking a time, none of these plain scheduling and info questions needed a person (hand-offs are tested separately), and first replies had a median of 4.6 s (p90 11.5 s). Rules-written replies took about 0.4 s, model replies about 7 s. Of 110 times offered, none were invented, busy or closed. 527 automated tests and about 23 review rounds by separate AI reviewer sessions (not human audits) back it. Honest limits: texts and emails are simulated. The calendar is a read-only Google Calendar feed; the app does not write to it. Sample business facts are placeholders. Demo follow-up timers run 60x faster. The owner phone alert is the one real outside action.