I build Pinwise, which uses the bring your own key setup this page describes, so I am not neutral. I have tried to write it so it is useful even if you never use my app. Nothing here is about what to take or how much. It is about where your data goes.
The two ways an app can give you an AI chat
Under the hood there are really only two setups.
- The app runs it for you. You type a question. It goes to the app company's server. That server adds your logs, sends the whole thing to an AI provider on the company's account, gets the answer back, and passes it to you. The company pays the AI bill, so it usually needs an account for you, a message cap, a subscription, or all three.
- You bring your own key. You sign up with an AI provider yourself and get an API key, which is basically a password tied to your own billing account. You paste it into the app. When you ask something, your phone talks to the provider directly. The app company is not in the middle and never sees the request.
Both can give you a good answer. The difference is the number of companies holding a copy of your question, and in a protocol log that question usually comes with your labs, your weight and your notes attached.
Who can see what
| App runs the AI | Your own key, cloud model | Your own key, local model | |
|---|---|---|---|
| App company | Sees every message and the logs sent with it | Sees nothing, no server in the path | Sees nothing |
| AI provider | Sees it, under the app company's account | Sees it, under your account and your settings | Nobody, it stays on your network |
| Who pays | The app, so it limits you | You, usually cents per conversation | You, in hardware and electricity |
| Account with the app | Usually required | Not needed | Not needed |
This is the general pattern. Any single app can do better or worse than its column, which is why the questions further down matter more than the label.
Your own key does not mean nobody sees it
I want to be straight about this, because it gets oversold. If you use a cloud model with your own key, the provider still receives what you send. Bring your own key removes the app company from the picture. It does not remove the AI company.
What it does change is whose rules apply. The data is handled under your agreement with the provider, not under some deal an app company signed that you never read. You can read those terms, change the settings on your account, delete the key, or move to a different provider. With the first setup you usually cannot do any of that.
Before you pick a provider, read its API data policy and look for two things: whether data sent through the API is used for training, and how long it is kept. Those answers change over time, so check the current page rather than trusting a blog post, including this one.
The local option
If you want nothing to leave your house, you can run a model yourself. Ollama is the common way to do it. It runs open models on a Mac or PC, and an app on your phone can talk to it over your home Wi-Fi.
The trade is real. Local models are smaller than the big cloud ones, your computer has to be on, and the first setup takes some fiddling. But it is the only version where the honest answer to "who sees my labs" is nobody.
What a key costs, roughly
API pricing is per amount of text, not per month. For the kind of use a tracker gets, a handful of questions about your own history a week, it tends to be small. Most providers let you set a monthly spending limit, and I would set one on day one so a bug or a lost phone can never run up a bill. Check the current prices on the provider's own page, since they move.
Questions to ask before you paste a key into any app
- Where is the key stored? On iPhone the right answer is the Keychain, the same place the phone keeps your saved passwords. Not a plain settings file, and never uploaded anywhere.
- Does the request go straight to the provider? Or does it pass through the app's server first? If there is a server in the middle, you are back to the first setup with extra steps.
- Does the app tell you exactly what it sends? Your question alone, or your question plus your labs, notes and history? You should be able to see the list before anything goes out.
- Does it ask again if you switch providers? Saying yes to one company is not saying yes to all of them.
- Can you take it back? Turning the AI off or deleting the key should be one tap.
- What does the app's privacy label say? Apple shows it on every App Store listing. If an app says it runs nothing on its side, the label should agree.
What the AI is for, and what it is not
Whichever setup you pick, an AI in a tracking app is good at one thing: reading your own records back to you. Summarizing what you logged last month, lining up a lab result with the dates around it, finding the note you wrote six weeks ago. It is not a clinician, and it should not stand in for one. Questions about what to take, or what a result means for you, belong with whoever ordered the test.
How Pinwise does it
I built Pinwise on the second and third setups because I did not want my own logs on anyone's server, mine included. There is no account and no Pinwise server at all, and the App Store privacy label reads Data Not Collected. The AI advisor works with Anthropic, OpenAI, xAI, or Ollama running locally, using your key, which is kept in the iOS Keychain. Before anything is sent, the app lists exactly what goes and to which provider, and asks. Switch providers and it asks again. Inventory counts and reminder times never leave the phone.
All the tracking is free. The advisor is part of Pro, which is $7.99 a month, $49.99 a year with a 7 day trial, or $129.99 once. It is iPhone only for now. If you are comparing apps, I wrote an honest comparison that includes the ones I compete with.