On-Device AI Is Like a Free Shooting Range: An Analysis of the LOCK TF IN Architecture
Sh0ny
24 июля 20262 min read
Updated 25 июля 2026In short
A fitness app featuring an AI “sergeant” takes an unconventional approach: the free version runs entirely on-device, with no accounts or tracking, while cloud storage and BYOK are available only in the Pro version. I’ll break down why this setup is appealing to AI app developers.
The LOCK TF IN app is a pocket-sized AI sergeant who scolds you for missing workouts. The idea itself isn’t new: gamification + a toxic coach. But the architectural solutions behind it are intriguing, and they’re more important than jokes about the couch and the alarm’s snooze button.
The free shooting range runs entirely on the device. No accounts, no tracking—nothing leaves your phone. AI dialogs are generated either locally or from a built-in library of responses. This isn’t just a marketing phrase—developer Brandon Hamilton has clearly drawn a clear line between on-device and cloud-based AI. Free = private and local; cloud = paid.
The Pro subscription adds a cloud-based AI coach with “adaptive roast responses,” weekly reports, and training analytics. But the most interesting detail is “bring your own AI key” for unlimited access. Users provide their own API key for the model and get unlimited interaction with the coach without additional charges for in-app requests. Subscriptions: $1.99/week, $4.99/month, $29.99/year.
This is a viable business model for niche AI apps with a small inference budget. The developer avoids the risk of unlimited cloud calls—passing the cost of tokens on to the user—and monetizes the packaging, UX, and analytics through the subscription. For audience-based apps with unpredictable traffic, this is more transparent than promising “unlimited AI” for $5 a month and then quietly cutting limits.
The app is 4.2 MB and requires iOS 17.0; there aren’t enough ratings yet to be displayed in the App Store. It’s rated 13+ due to profanity—there’s a Clean Mode, but the developer honestly warns that “the coach hits harder without censorship.”
The main takeaway isn’t about fitness. If you’re building an AI app, think about this same boundary: the on-device model as a free, private shooting range; the cloud as a premium option; and BYOK as an escape valve for heavy users. This reduces your risk regarding inference costs and gives users a choice about what to pay for.
Source: Hacker News - Newest: “AI” “LLM”