Agxntsix built Vic, a production voice AI receptionist for Victoria's Beauty & Aesthetics, an aesthetics spa in the Bronx, New York. Running on Vapi and GoHighLevel, Vic answers every inbound call in English, Spanish, or Russian and completes real work mid-call — checking availability, booking, rescheduling, and canceling appointments through 11 live tool-call functions — instead of taking messages.
Agxntsix Team
Updated on Jun 2026
Victoria's Beauty & Aesthetics is a medical-aesthetics spa in the Bronx, New York, offering lash extensions, microblading, ombre brows, and lip treatments to clients who speak English, Spanish, and Russian. In December 2025 the spa engaged Agxntsix to take over its phones. Agxntsix built Vic — a voice AI receptionist on Vapi, integrated with the spa's GoHighLevel CRM and calendar — backed by a Bun + Hono orchestration layer on Railway, Neon Postgres call records, and an automated KPI pipeline. Onboarding December 18, 2025; production hardening through March 2026: 195 commits, 23 shipped features.
For years, the owner's personal cell phone was the business's main number, and she could not get to every caller. Missed-call recovery meant texting a screenshot of her last ten calls to virtual assistants, who dialed each number back by hand. Spam buried real bookings, reminders were manual, and every reschedule required a callback — each missed connection a lost appointment.
The workflow captured in the December 18, 2025 onboarding session was brutally manual: the owner, busy performing treatments, screenshotted her missed-call log and texted it to virtual assistants, who dialed each number back one at a time. Follow-up requests sat with a chatbot that stopped and waited for a VA to apply a tag by hand. With services spanning the spa’s full menu, from quick fill-ins to premium multi-hour sessions, every call that hit voicemail was revenue walking to the next spa.
Agxntsix engineers voice AI as production software, not a configured chatbot: custom code under version control, a 25-feature roadmap gated on per-feature test criteria, Playwright end-to-end tests, and an automated pipeline that scores every production call against client-specific KPI thresholds — a receptionist that is measured, monitored, and improved continuously.
Agxntsix deployed Vic, a trilingual AI receptionist that answers Victoria's inbound line and completes actions while the caller is still on the phone. Every call triggers a GoHighLevel lookup inside Vapi's webhook window: returning clients are greeted by first name, their upcoming appointment referenced in the opening line. Mid-conversation, Vic invokes 11 live tool-call functions — checking real calendar availability, booking, rescheduling, and canceling appointments, scheduling AI callbacks, texting directions via SMS, and creating or updating CRM contacts. Each tool response carries behavioral guidance for the model ('Offer conversationally. Confirm before booking.'), keeping the agent on-script without adding latency. A companion outbound agent handles reminders. Every call is saved to Neon Postgres and scored by the KPI pipeline.
The onboarding call and the first commit happened the same day — December 18, 2025. One day later the system was live on Railway with end-to-end tests passing. Work was tracked as a 25-feature roadmap, each feature gated on explicit test criteria; 23 shipped. Agent prompts and tool definitions live as files in the repository and sync to Vapi via scripts, so every prompt change is reviewed and reversible. Across 195 commits through March 2026, Agxntsix added the outbound reminder agent, KPI pipeline, Slack alerting, and analytics dashboard.
The centerpiece is that Vic acts instead of taking messages. Most AI phone agents end calls with 'someone will get back to you'; Vic ends calls with the appointment on the calendar — auto-creating CRM contacts for first-time callers and auto-canceling pending reminders when an inbound caller resolves their appointment. The caller-ID greeting is the differentiator clients notice: returning callers hear their name and upcoming appointment before they say a word.
The headline result is structural: calls now end with the work done. A caller hears real availability, confirms a time, and is booked into GoHighLevel before hanging up — no screenshot, no VA callback. First commit to live, tested deployment took two days, and every production call since has been logged, scored, and trended.
| Metric | Before | After |
|---|---|---|
| Answering the phone | Owner's personal cell was the main line; she missed many calls | Vic answers in English, Spanish, or Russian |
| Missed-call recovery | Call-log screenshots texted to VAs for manual dial-backs | Calls handled live; unfinished conversations get a scheduled AI callback |
| Booking an appointment | Required a human callback to check the calendar | Booked mid-call against the live GoHighLevel calendar |
| Returning clients | Every caller treated as unknown | Caller-ID lookup greets clients by name with their upcoming appointment |
| Appointment reminders | Manual texts and VA dial-downs of the appointment list | Outbound AI reminder agent with voicemail detection and outcome tracking |
| Call quality oversight | No visibility into what happened on calls | KPI pipeline scores every call daily; Slack alerts on errors |
Yes. Vic, the AI receptionist Agxntsix built for Victoria's Beauty & Aesthetics, executes 11 live tool-call functions during the conversation — checkAvailability, bookAppointment, rescheduleAppointment, cancelAppointment, scheduleCallback, sendDirectionsSms, getCustomerAppointments, createContact, getContact, and updateContact — writing directly to the GoHighLevel calendar and CRM, so the appointment is confirmed before the caller hangs up.
Each tool is one focused capability invoked through Vapi's tool-calls webhook. The orchestration layer executes the request against the GoHighLevel v2 API and returns data plus procedural guidance — available slots come back with the instruction to offer them conversationally and confirm before booking — keeping behavior consistent without bloating the prompt.
Every inbound call triggers a real-time GoHighLevel CRM lookup on the caller's phone number. Returning clients are greeted by first name, and if they have an upcoming appointment, Vic opens with the service and date on file — asking whether they're calling about that visit. New callers get a standard greeting and a contact record created mid-call.
This happens inside Vapi's assistant-request webhook, which must respond within 7.5 seconds. Agxntsix's Bun + Hono server performs the lookup, assembles caller context, and pre-renders a personalized first message before the agent speaks — turning the identification dance into an instant, personal greeting.
Vic auto-detects the caller's language from their first utterance and locks into English, Spanish, or Russian for the entire call, switching only when the caller does. The prompt specifies culturally calibrated tone — warm and familiar in Spanish, slightly more formal in Russian — which matters in the Bronx, where Victoria's client base spans all three languages.
Language handling is specified in the version-controlled system prompt: detect from the first utterance, respond in kind, maintain for the full call. Because every booking flows through the same tool-call layer regardless of language, a Russian-speaking client gets the identical mid-call booking experience an English speaker does.
The platform is templated for multi-client deployment: a YAML config system defines each client's industry, call types, and KPI thresholds, with documented templates for Healthcare, Dental, Legal, and Sales verticals. Scoped roadmap items include speed-to-lead callbacks for Facebook leads, no-show win-back calls, and database reactivation campaigns.
Agxntsix took Victoria's Beauty & Aesthetics from first commit to a live Railway deployment with passing end-to-end tests in two days — December 18 to 19, 2025. Production hardening continued from there: caller personalization, eleven tool-call functions, trilingual support, an outbound reminder agent, and a KPI monitoring pipeline shipped across 195 commits through March 2026.
Vic speaks English, Spanish, and Russian. It identifies the caller's language from their first words, responds in that language, and maintains it for the full call — following the caller's lead if they switch. Callers in unsupported languages are offered a choice of the three. Tone is calibrated per language: casual English, warm Spanish, more formal Russian.
Agxntsix built an automated analysis pipeline that pulls Vapi call logs, scores each call against client-configurable KPI thresholds — success rate, error rate, echo rate, cost — and tracks trends across dated history snapshots. Analysis runs incrementally on new calls via a daily midnight job, and production errors trigger Slack notifications to the engineering team.
Yes. In January 2026 Agxntsix added an outbound reminder agent that calls clients ahead of appointments, detects voicemail, and records confirm, reschedule, or cancel outcomes through server-side decision endpoints with retries. When a client calls back in and resolves their appointment, pending reminders for that appointment are canceled automatically.
Yes. The platform is client-agnostic by design: each deployment is a YAML configuration defining the client's industry, call types, cost model, and KPI thresholds. The config system ships with a documented template covering Healthcare, Dental, Legal, and Sales verticals, so the same Vapi-plus-CRM architecture redeploys for a new business without rebuilding the orchestration layer.
Agxntsix builds production voice AI that answers in your clients' language and completes the booking before they hang up. Book a consultation to scope your deployment.