A transit team is configuring a rider-assistant inside a Foundry project. Which setup step is essential?

Configuring a rider-assistant inside a Foundry project requires configuring the model deployment and the agent deployment the assistant will use. Those deployments are the runtime building blocks the app consumes: the model provides generative capability; the agent deployment encodes how that capability is applied with tools, knowledge, or instructions. Skipping them leaves a project shell without a working assistant.

Configuring a storage account in the Foundry project to retain rider-assistant chat history for detailed review fails because chat-history storage alone does not provide the model deployment and agent deployment the assistant requires to answer riders. Configuring an application connection in the Foundry project to show rider-assistant replies in the transit app fails because wiring the app display does not configure the model deployment and agent deployment that power the assistant itself. Configuring the model deployment and reusing a shared agent deployment from another Foundry project for rider queries fails because the rider-assistant requires an agent deployment configured in its own Foundry project, so borrowing one from another project does not provide the project-local deployment the assistant will use.

Exam caveat: connection strings, managed identity, and environment separation still matter once deployments exist.

Operational check: from the rider app, call the configured agent/model endpoint successfully in a non-prod project first, confirm identity is least-privilege, and verify no raw database admin secrets are present in client sessions.