Who funds AI requests inside a generated app?
A direct answer, the limits that matter, and a practical way to check the workflow.
Direct answer
Managed Uvilno.ai requests consume the creator's studio allowance, subject to visitor, IP, project and owner admission limits. Each response is capped at 600 output tokens. An AI feature therefore needs a product scope and failure states; it is not an unlimited model endpoint for visitors.
45-word answer · Reviewed 2026-10-27Separate the builder's generation request from the app's later inference. A generated AI helper can continue making permitted requests when real visitors use it, so its operating budget matters after the frontend is saved. Never describe those calls as free simply because the visitor did not paste an API key.
Design short outputs, restrict the feature to its intended job and preserve input when a request is denied. Avoid storing sensitive prompts unnecessarily, and validate structured results before saving them. Usage limits protect the service but do not replace a thoughtful product boundary.
What to verify.
| Area | Documented behavior | Your verification |
|---|---|---|
| Owner funding | Actual measured app inference uses creator capacity. | Review the app feature and remaining allowance together. |
| Response bound | Managed output is capped at 600 tokens. | Design a short response rather than a long report. |
| Admission | Several identity and project limits apply. | Exercise a safe denied-request state without retry loops. |
Drag the slider, use the arrows, or scroll within the table.
Try the workflow yourself.
Use this checklist in your own disposable project. These marks record your review; they do not run tests or certify an application.
0 of 3 review steps marked by you.
Read the editable test brief
Plan a concise AI writing-feedback tool for signed-in app visitors. Use the actual managed inference SDK, request a short bounded response and show allowance-denied and provider-failure states. Keep the original text available, avoid automatic retry loops and explain that the creator funds app inference separately from source ownership.
Preparing a brief does not start generation. Review and send it explicitly; builds use the normal allowance, and source access requires entitlement.
Sources and scope.
Published by the Uvilno editorial team on . This is product documentation and editorial guidance, not an independent ranking or a promise that every application will pass. Vendor plans can change; official sources establish their own terms.
Does opening this test use my allowance?
Preparing the test brief does not invoke a model or create a billable build. Review the draft and explicitly send it to generate; normal usage and source entitlement rules apply.
Check the architecture. Compare the tradeoffs.
Follow the data model, account boundary and source handoff before choosing a builder.
Compare builders
Plan a full-stack app
Review the evidence
- Full-stack builder buyer’s guide
- Four full-stack workflows compared
- Plan data, sign-in and checkout
- Machine-readable evidence and methodology
- Remixable app blueprints
- Direct answers to builder questions
- Free planning and migration tools
- 108 full-stack engineering prompts
- Public SDK documentation
- Recorded checks and open evaluation protocol
- RSS updates for guides and comparisons