Coaching
BuiltClaude drafts multi-week programmes from each member's goals, level and equipment, and revises them on a coach's feedback.
A coach approves every programme.
AI-native gym software · Built in Kampala
The AI-native gym platform.
Fitness Arena runs a gym end to end — members, assessments, coaching, classes, staff pay and the cash desk — with Claude built into the core. Today it drafts training programmes; next it explains the finances and helps keep members coming back.
In active developmentClaude in the coreAI drafts · people decide
MonBack squat · Romanian deadlift · Plank4 × 6
WedBench press · Seated row · Farmer carry4 × 8
FriDeadlift · Split squat · Rower intervals3 × 5
Ask for changesApprove for member
The problem
Gyms run on paper, cash and chat
Most gyms we know track members in notebooks, take payment in cash and mobile money, and coordinate staff over WhatsApp. The result is the same everywhere:
Progress is a guess
Members can't see whether they're improving, so they drift away. Coaches write plans from memory, not data.
Money leaks
Cash at the desk, informal advances and per-class coach pay settled by hand — with no record when amounts disagree.
Coaching doesn't scale
A good coach can write a few personal programmes a week. Every member deserves one.
AI-native
Claude across the platform
We're designing the platform with Claude in the loop — not a chatbot bolted on, but a layer that drafts, explains and recommends inside the workflows staff already use, with a person making every decision that matters.
AI foundation
BuiltOne layer every Claude feature runs through: model settings, spend limits, usage tracking, approved-model rules for health data and an audit log.
New AI features inherit the guardrails.
Finance assistant
PlannedClaude explains the month to an owner — revenue, what's outstanding, what needs chasing — and ranks anomalies into a worklist.
The model never produces a number. Every figure is computed in SQL and cited.
Member growth assistant
PlannedClaude drafts win-back and renewal messages, proposes rewards for milestones, and answers members' questions on WhatsApp from the gym's own knowledge base.
The model drafts; a person sends. Rewards come from an approved list, and anything account-specific goes to staff.
AI drafts, people decide
No programme, message, reward or post reaches a member without a person approving it.
Facts come from the database
Numbers, segments and eligibility are computed; Claude explains and writes, it doesn't invent.
Minimum data, approved models
Each call gets only what it needs, with consent where health data is involved.
Built with Claude, too
We build the platform with Claude Code. Since July 2026 it has co-authored most of our changes — 277 of the 391 commits in the codebase — with every change tested in CI before it ships.
~70%of commits co-authored with Claude Code
The platform
One system for the whole gym
A staff dashboard and a member app on one platform. Everything below is built; we're hardening it with Fitness Arena's coaches and front desk ahead of launch.
Assessments & progress
BuiltFitness tests with defined protocols, and the analytics on top of them.
- Progress over time per test
- Percentiles against the gym's population
- Personal records
Programmes
BuiltA programme builder for coaches and a training log for members.
- Multi-week plans, assigned per member
- Today's workout in the member app
- Set-by-set logging
AI programme drafting
ClaudeClaude drafts a programme from the gym's own exercise catalogue. A coach reviews, refines and approves it.
- Draft only — never sent without approval
- Refine from coach feedback
- Streams as it drafts
Member-requested programmes
BuiltMembers ask for a programme around their goals; it lands in a coach's queue.
- Explicit, withdrawable consent
- Baseline tests before generation
- Midpoint and end retests vs baseline
Group classes
BuiltTimetables, per-session enrolment and attendance that feeds coach pay.
- Attendance confirmed by a second person
- Corrections need a separate decider
- Coach payout reporting
Staff pay & payroll
BuiltSalaries, class payouts and advances, with the owner as the approver.
- Owner-only approval of pay
- Payees acknowledge or dispute
- Mobile-money payouts via Eversend
Money desk
BuiltThe front-desk till, spending and money in — where cash gyms lose track.
- Open, count and close the till
- Expenditures and collections
- Banking the till, owner-approved
Equipment
BuiltInventory, service records and schedules for every unit on the floor.
- Status changes are audited
- Out-of-service kit is excluded from AI drafts
- Coach-set exercise substitutions
Members & plans
BuiltMembers, plans, subscriptions and payments, behind role-based access.
- Owner, admin, staff, finance and member roles
- Member data export and anonymisation
- A staff dashboard and a member app
Claude in coaching · Built
A coach's first draft, by Claude
Claude, through the Anthropic API, turns a member's goals, level and available equipment into a structured multi-week programme. It drafts; a coach decides. Here is the path every programme takes.
Request
A coach starts a draft, or a member asks for a programme and gives consent.
Minimise
Only what the draft needs is sent — goals, level, equipment. No names, no emails.
Claude drafts
Claude builds the week, choosing exercises only from the gym's catalogue.
Validate
Every exercise is checked against the real catalogue; incomplete output is rejected.
Coach approves
A coach edits, asks for a revision, or approves. Only then does a member see it.
Spend limits before every call
A monthly USD cap, a per-person daily limit and a concurrency cap. Owners are emailed as the cap nears.
Health data stays on approved models
Member health inputs only go to models an admin has approved with zero or short data retention.
An audit log for every generation
Generations, refusals and model-setting changes are recorded in an append-only log.
Drafts don't linger
The content of drafts nobody accepted is wiped after a retention window.
Grounded in what the gym has
Drafts use the gym's in-service equipment — or a member's home kit — so plans are doable.
Learning from coaches
When coaches edit an accepted draft, we record how much changed to measure draft quality.
1Baseline tests recordedDone
2Claude draft · catalogue-checkedReady
3Coach review & approvalNext
4Retest at the midpoint, compared to baselinePlanned
Illustration of the request flow — not live member data.
Built for how gyms here pay
Cash, mobile money and two pairs of eyes
Payments in Uganda move through cash and mobile money, in UGX. The platform is designed around that, with integrations with Eversend for mobile-money payouts and collections — and a rule that the person who records money is never the person who approves it.
- Pay needs the ownerWhoever prepares a payment can't approve it, and nobody approves their own pay.
- Attendance needs a second personCoaches submit class attendance; an admin confirms it before it counts towards pay.
- Banking the till needs approvalCash banked by deposit counts only once an owner who didn't record it approves it.
- Everything leaves a trailPay, class and AI events are kept in append-only audit records.
Engineering
Serious foundations
A Go API on PostgreSQL with a React front end. Every change runs the Go test suite against a real, fully migrated database in CI before it can ship, and new capabilities go live behind feature flags.
390+
commits since August 2024
~200
API endpoints
277
commits co-authored with Claude Code
70+
Go test files, run on real Postgres
GoPostgreSQLReactViteTailwindAnthropic API · ClaudeEversendGitHub ActionsCloudflare
Figures from our code repository, October 2026.
Also next
Beyond the AI lanes
Designed and specified, not yet built.
Planned
Wearables
Bringing heart rate and activity from members' devices into their progress.
Planned
Public gym site & shop
Live plans, the class timetable, enquiries and click-and-collect merchandise.
Built with a real gym
Fitness Arena, Kampala
We're building the platform alongside Fitness Arena, a gym in Kampala, where it will launch first. Its coaches and front desk shape every workflow — from how a till is closed to how a programme is approved — so the software fits how a gym actually runs.
Test.Train.Retest.
Contact
Let's talk
Gym owners, coaches and partners — we'd like to hear from you.