Troubleshooting & Support

Work through the relevant section below before contacting support. Most issues have a straightforward cause that you can confirm and resolve in a few minutes.

Calls are not appearing in the dashboard

Work through this checklist in order:

1. Confirm payi_instrument() runs before client initialization. Pay-i patches the OpenAI client at instrumentation time. If a client is created before payi_instrument() is called, that client is not instrumented and calls will not be tracked. Move payi_instrument() to the top of your startup sequence, before any client is created.

2. Check for client-level errors. If your application is throwing exceptions during inference calls, the request may not be reaching Pay-i. Review your application logs for connection errors, authentication failures, or timeouts.

Requests show as unknown model

Requests appear in the dashboard but are attributed as unknown model when the endpoint name Pay-i received is not yet configured for your Application.

What to do: Unlike Azure AI Foundry, where model mappings associate a specific deployment name with a model, Databricks host_mappings is purely client-side configuration. Verify that your Databricks URL and category values are correct. If they are, contact your Pay-i customer success representative with the endpoint name shown in the dashboard—Pay-i configures the resource so future requests are attributed and priced correctly. Requests captured while the model is unknown are not retroactively updated, but data collection continues without interruption.

Host mapping mismatches

If requests are appearing in the dashboard but attributed to the wrong category, the most likely cause is a host mapping mismatch.

The Pay-i SDK normalizes host strings automatically—it strips the scheme, any path, and trailing slashes, and lowercases the result. What matters is that the value you pass to host_mappings is the same value you pass to your Databricks client. Reading DATABRICKS_HOST from the environment in both places makes this correct by construction:

host = os.getenv("DATABRICKS_HOST")
payi_instrument(config={"databricks": {"host_mappings": {host: {"price_as_category": "system.databricks"}}}})
client = OpenAI(api_key="<YOUR_DATABRICKS_TOKEN>", base_url=f"{host}/serving-endpoints")

If those values differ, Pay-i falls back to a default category—no error is raised.

Support

For questions or issues with Pay-i instrumentation for Databricks, contact the Pay-i team:

• Email: [email protected]
• Customer success: Contact your assigned customer success representative directly

For Databricks Model Serving documentation, visit docs.databricks.com/machine-learning/model-serving.

All product names, logos, and trademarks referenced in this documentation are the property of their respective owners.