Skip to main content

AI per App

Each app (project) chooses who serves its AI model:

ModeWho pays the AI providerSetup
Your own key (BYOK)You, directly to OpenAI, Anthropic, or another providerAdd a provider and its API key to the project, then pick a model. See Recommended models.
AI included (managed)Appilots. The app's plan includes a monthly number of missions.Nothing. Appilots serves the model with its own key.

The choice is per app, so one workspace can run some apps on its own key and others on AI included. Change it under Workspace → Billing → AI per app.

Availability

AI included opens per deployment. While it isn't available, the dashboard only offers Your own key, and every app runs on BYOK. Apps created before per-app modes existed run on BYOK.

Your own key (BYOK)​

The agent uses the provider key configured on the project, and the project's budget settings apply. Appilots never falls back to its own key for a BYOK app.

There's no monthly mission cap in this mode.

AI included (managed)​

  • It requires a paid plan. On a workspace without one, the app gets a clear "requires a paid plan" error instead of quietly switching to another key. Subscribe, or switch the app to your own key.
  • The included missions are a guide, not a wall. The dashboard shows "N of M included missions this month". Above that number, the assistant keeps working and nothing extra is charged. If an app stays above its included missions for three consecutive months, it moves to the next tier at renewal. The dashboard counts those months as they happen.
  • The model is GPT-6 Luna, picked by Appilots. Your project's provider and model settings don't apply in this mode.

A mission counts when it completes and the server can verify the work landed. Missions run with a test key, sandbox runs, and evals never count.

Last updated on