Athlo

The Problem

Try getting a straight answer out of a big-box gym chain. The phone queue runs an hour or more, the support email has quietly been switched off, and a simple question — can I freeze my membership over the summer? — turns into a week of being bounced around. It's not that the staff don't care; it's that front-desk teams are buried under the same dozen questions, arriving across email, Instagram DMs and WhatsApp, with no system to triage them. I lived this as a member, and saw the same backlog from the gyms' side: a customer-service channel that has effectively collapsed.

The Solution

Athlo is a SaaS product I sell to gyms that automates their customer-service desk end to end. It pulls every channel — email, Instagram and WhatsApp — into one inbox, and for each incoming message an AI drafts a reply in the gym's own voice, grounded in that gym's facts. The owner skims the draft and sends it with a single keystroke, or edits it first. It's deliberately draft-and-approve, not fully autonomous: the gym keeps its hand on the wheel while the hour of repetitive typing disappears.

How the AI Works

There is no queue broker and no polling — Postgres is the spine. When a message lands, a webhook writes it to the database, and an INSERT trigger fires a NOTIFY. A long-lived worker is LISTEN-ing, picks it up in under a second, builds the prompt, calls the model, and writes back a ready draft. The whole path from member email to reviewable answer takes a few seconds.

Sounding Like the Gym

A generic chatbot answer is worse than no answer — members can smell it. So the prompt isn't one monolith; it's composed from small, ordered skill modules. Three are mandatory and can never be switched off: role (who the assistant is), company-info (the gym's knowledge base, which grounds every fact), and safety (injection fencing, no invented prices, no legal or medical advice). Around them sit optional modules a gym can tune — tone, banned words, sign-off — and the highest-leverage piece: a handful of real voice examplesthe model imitates. Inbound text is fenced as data, never instructions, so a member can't talk the assistant out of its rules.

It Learns from a Skip

When the AI doesn't know something, the owner doesn't just fix the one reply — they teach the system. Hitting Skippaon a draft and flagging a knowledge gap opens an inline box to add the missing fact to the gym's knowledge base. The draft then regenerates on the spot, now grounded in the new answer — no page reload, no leaving the inbox. Every gap closed this way makes the next hundred answers better.

The Stack

The whole product is a small, boring-on-purpose set of processes with Postgres at the centre. The web app is Next.js 16 / React 19 (Server Components, TypeScript strict), which renders the three-pane inbox and hosts every webhook and route handler. Supabase Postgres in the EU is the spine: multi-tenant isolation is enforced in the database with row-level security keyed on gym_id, OAuth tokens live encrypted in Vault, and an append-only audit log records every send. A single Bun worker does the drafting against Claude Sonnet; Stripe handles subscriptions; a daily cron keeps the mailbox connections alive. No Kubernetes, no Redis, no queue — and everything stays in EU regions by design.

Where It Stands

Athlo is live at athlo.se and running with pilot gyms. The bet behind it is simple: members deserve a real answer in minutes, not a week, and gyms shouldn't have to choose between hiring more front-desk staff and letting their inbox rot. An owner-approved AI draft for every message — across every channel, in the gym's own voice — turns a collapsed support channel back into a working one. Built solo, end to end: product, AI, full stack, and the EU-compliant plumbing underneath.