Playbook · Building with AI

Updated

How do you add AI features to a mobile app?

To add AI features to a mobile app, first decide where the model runs. On-device models from Apple and Google have no per-call bill and work offline, but only on supported phones. Cloud models work on every phone and are stronger, but cost money per call, need a network, and send user data to a third party you must disclose.

Apple’s App Review Guidelines say: “You must clearly disclose where personal data will be shared with third parties, including with third-party AI, and obtain explicit permission before doing so.”
Source: Apple, App Review Guidelines 5.1.2(i), read 17 Nov 2026

From my work

For Groupe Rougier I built a React Native app for lumberjacks to enter data on the go, with barcode scanning. The data went into a local SQLite database on the phone, and I built the sync with the company’s existing Odoo ERP.

A cloud AI feature needs the same kind of thinking about the network: decide what the app does when the call can’t go through. EZ Chat, my own app live on the Overwolf store, translates chat between players with the OpenAI API, with paid subscriptions through Tebex. It runs on desktop, not on phones, but it is a live AI feature people pay for.

Artificial intelligence in mobile apps: on-device or cloud

  • iOS: Apple’s Foundation Models framework, from iOS 26, gives apps access to the on-device and Private Cloud Compute models designed for Apple Intelligence. Apple lists summarization, entity extraction, text and image understanding, and refinement as strengths of the on-device model. It needs a device that supports Apple Intelligence.
  • Android: Gemini Nano runs in Android’s AICore system service. Google says it works “without needing a network connection or sending data to the cloud.” The ML Kit GenAI APIs cover summarization, proofreading, rewriting, image description and speech recognition, among others. Google’s own docs also point you to a cloud fallback for phones where Gemini Nano is unavailable or unsupported.
  • Cloud: OpenAI, Anthropic or Google models through your own server. Works on every phone, paid per token. In OpenAI’s gpt-6 family, prices run from gpt-6-luna at $0.10 in and $0.50 out per 1M tokens to gpt-6-astra at $10.00 in and $50.00 out (seen 17 Nov 2026).

React Native AI apps

React Native gives you one code base for both stores. The on-device models are native APIs, though: Foundation Models is a Swift framework, and the ML Kit GenAI APIs are Android libraries. So a React Native app either needs a small native module per platform, your own or a library, or uses the cloud path on both. You can also use both, on-device where it exists and cloud as the fallback. More code, and the option I would pick.

Never ship the API key in the app

Anything inside the app bundle can be pulled out of it, and somebody with a free afternoon will. Call the cloud model through your own server, which checks who the user is, applies a limit per user, and logs the cost. This is also where you strip personal data you don’t need to send.

What the stores require

  • Apple, guideline 5.1.2(i): disclose where personal data goes to third-party AI, and get explicit permission first.
  • Google Play: “Apps that generate content using AI must contain in-app user reporting or flagging features,” and must prevent restricted content.
  • For your privacy policy: OpenAI says data sent to its API is not used to train its models unless you opt in, and abuse monitoring logs are kept for up to 30 days. Zero data retention needs OpenAI’s prior approval.

Design for no signal and for wrong answers

  • Show AI output as a suggestion the user can edit, not a final answer.
  • Queue requests when the phone is offline, or hide the feature until it can work. Real phones go into lifts and basements. The demo phone never does.
  • Set a timeout and show a clear message when the model is slow. A spinner that never ends does not count as one.
  • Keep the app useful when the model is unavailable.

When a mobile feature should not use AI

Barcode scanning, form checks and search over a short list don’t need a language model. A barcode has one right answer, and plain code finds it faster, offline, the same way every time. Use AI where the input is messy text, speech or images and a close answer is still useful.

What it costs

I sell this, so read it with that in mind. I build MVPs on the Full plan, $4,995 a month, starting with one user type and one core flow. The build is split into tasks, and the number of months depends on scope, agreed on the kickoff call.

If you only need one AI feature added to an app you already have, it is done on a monthly plan, one task at a time: Limited at $1,995 a month or Full at $4,995. Both are a flat monthly price, pause or cancel anytime. On the Full plan it usually ships within the first month.

Read next

Sources

Start

Find out what stands between you and launch.

Start the 14-day Trial: your problems ranked in the first 5 business days, then fixes for what fits in the 14 days. Not ready yet? Ask for a free teardown.

Start the 14-day Trial · $995

Secure checkout via Stripe · Refund policy

Founding rate: the first 3 clients get 20% off, in return for a named testimonial and a case study. On a monthly plan, the 20% applies to the first month.

Prefer to talk first? Book 30 min or email badii@badii.studio

Free teardown

48h · max 5 / week

A 5 to 10 min recorded review of your live app's public surface. No access needed, just the URL.

I use these details only to reply and record your teardown, and delete them after 12 months. Privacy