migrate-from-model-serving

Migrate your MLflow agent from Databricks Model Serving to Databricks Apps by downloading artifacts, relocating /code and /artifacts directories into an app folder, and updating imports to match the new deployment structure.

Databricks Apps deployment converts predict and predict_stream functions into decorated @invoke and @stream endpoints, supporting synchronous execution and optional asynchronous conversion for reliable agent operation.

Configure Databricks Asset Bundles for an MLflow ResponsesAgent by mapping the resources section from your MLmodel file directly into the databricks.yml configuration to achieve production-ready deployment.

Yes, you can deploy an MLflow ResponsesAgent asynchronously using Databricks Apps by enabling async conversion during migration to support decorated @invoke and @stream endpoint workflows.

The best way to move MLflow code dependencies is to download original Model Serving artifacts, extract the /code directory, copy it into the new app directory, and update all necessary import paths.

Your MLflow agent needs @invoke and @stream decorators in Databricks Apps because the target server requires this specific interface to replace predict and predict_stream methods used in Model Serving.