On 54 the AI of an app runs on the server with the credit your plan includes, so there is no API key for a visitor to steal; the app saves its data in records in your site, and visitors log in with a code to their mail, Google or a password.
Putting a model inside an app is the easy half. The hard half is **everything around it**: where the key lives so nobody can take it, where the answers get stored, who is allowed to ask, and what it costs when a hundred people use it on the same afternoon.
Swipe to compare all columns →
| Wiring it yourself | 54 | |
|---|---|---|
| The API key | In your code, or in an env var on a server you maintain | Never in the page: the call runs on the server with the credit your plan includes |
| What a use costs | Your own bill, per token, after the fact | Against the AI your plan includes, with a daily ceiling per account |
| Where answers are stored | A database you set up | Records in your site, visible in the admin |
| Who can ask | An auth system you build | A code to their mail, Google or a password, already there |
| Shipping a change | Commit, build, deploy, hope | Say what you want changed and publish |
Written from each product's public documentation. Products change: check theirs before you decide. No prices here for the same reason.
No. The call runs on the server against the AI included in your plan.
The platform picks from a ladder and falls back if one is down, so a provider having a bad day does not take your app with it.
The app says so instead of failing silently, and you can top up.
There is a daily ceiling per account and per site, checked before the model is called.
In records attached to your site. Public records are readable by anyone; the rest need a session.
Yes. That is a line in what you ask for.
Yes, images and documents.
Yes — a site assistant with its own prompt and its own knowledge.
Yes, from the admin.
Yes, and that is included in the free plan.