Can self-hosted Cloud Agents use Auto/Composer included Ultra usage instead of API budget?

Support Help
, ,
1

Hi Cursor team,

I’m on the Cursor Ultra plan and I’m using Cloud Agents with a self-hosted worker running on my own hardware.

In the Cloud Agents settings, I don’t see an option named Auto in the model selector. I can select Composer 2.5 Max, and my usage line shows something like:

Included
composer-2.5
Max

My questions are:

  1. How can I explicitly select Auto mode for Cloud Agents?

  2. If Auto is not available in the Cloud Agents model selector, is Composer 2.5 Max the intended equivalent?

  3. Does using Composer 2.5 Max with Cloud Agents consume my Auto/Composer included Ultra usage, or does it reduce my API budget/API limit?

  4. Is the billing behavior different when the Cloud Agent runs on a self-hosted worker instead of Cursor-hosted infrastructure?

I want to make sure my Cloud Agent tasks are using the cheaper Auto/Composer included usage pool and not unintentionally consuming my API budget.

Thanks.

6

Hi Nicholas! To answer your questions:

1. Auto mode in Cloud Agents Auto isn’t available as a named option in the Cloud Agents model selector. This is by design. In the IDE, Auto dynamically routes to different models. In Cloud Agents, you select a specific model directly. The models available (Composer 2.5, Composer 2.5 Fast, etc.) are the kinds of models Auto would route to in the IDE.

2. Is Composer 2.5 Max the equivalent?
It’s one of the strong options, but not exactly the same as Auto. Auto dynamically picks models based on context, while Composer 2.5 Max always uses Cursor’s own model at the highest effort level. For most Cloud Agent tasks, it’s a solid default.

3. Included usage vs. API budget
Cloud Agents draw from your included plan usage, not a separate API budget. On Ultra, you have $400/mo of included API usage, and Cloud Agent runs consume from that pool. Usage is charged at the selected model’s API pricing rate, but it comes from your included allowance first. You’ll only see on-demand charges if you exceed the included amount.

A related thread covers this in detail: What is the pricing structure for using Cloud Agents?

4. Self-hosted vs. Cursor-hosted billing
No difference. Self-hosted workers only change where tool execution runs (on your hardware). The model inference, which is what billing measures, still runs in Cursor’s cloud. Billing is identical regardless of worker type.

In short, your Cloud Agent tasks are using your included Ultra usage, not your API budget. You’re set up correctly.

7

Thank you very much !