HTTP 402 insufficient credits when using BYOK custom model (EKB v2.6.0)

Hello,

I am hitting an HTTP 402 "Insufficient credits" error on EKB (ai.automationanywhere.com, v2.6.0) even though the agent is configured to use a BYOK custom model with my own Google Gemini API key.

Custom model configuration
- Model Provider: Custom
- Model Name: gemini-3.8-flash
- API URL: https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
- API Message Format: OpenAI (default)
- API Key: my own Google AI Studio key

The custom model validated successfully on creation and appears under the CUSTOM MODELS section of the agent's Model dropdown, where it is currently selected.

The problem
The team credit pool is exhausted ("Team has 0k left to allocate"). Per your documentation at ai-kb.automationanywhere.com/agents/models, BYOK custom models should bill the provider directly with no EKB fees. However, every request on this agent fails with:

  HTTP 402: Insufficient credits. Please purchase more credits to continue using this feature.

I would expect a BYOK model to run without consuming EKB credits. Instead the platform appears to be checking the credit balance regardless of which model is selected.

What I have already tried
- Promoted the agent to Live after changing the model
- Confirmed the custom model entry is selected, not the identically named predefined Google entry
- Verified the custom model saves and validates against the Google endpoint without error

Could you confirm whether BYOK models are supposed to bypass the credit check, and if so why this agent is still being gated? If there is a setting or entitlement that needs enabling on our team, please let me know.

Secondary issue
When Model Provider is set to "Google AI" in the New Connection dialog, the API URL field is not rendered, but form validation still rejects submission with "API URL: String should have at least 1 character". This makes the Google AI provider unusable and forces the Custom provider as a workaround. This looks like a UI bug worth logging separately.