migrate-from-model-serving

To migrate an MLflow ResponsesAgent from Databricks Model Serving to Databricks Apps, you extract the original code and artifacts, transform the predict() and predict_stream() methods into @invoke and @stream decorated functions, and deploy using Databricks Asset Bundles.

You convert ResponsesAgent predict() and predict_stream() methods by transforming them into decorated functions using @invoke and @stream, supporting either synchronous or asynchronous execution paths.

Yes, the migration process preserves and migrates code dependencies, prompts, tools, and memory configuration from the original ResponsesAgent when deploying to Databricks Apps.

Yes, migrating to Databricks Apps supports both synchronous and asynchronous paths when converting the ResponsesAgent methods to @stream and @invoke decorated functions.

The best way to deploy a migrated MLflow agent is by packaging the transformed @invoke and @stream decorated functions and deploying them end-to-end via Databricks Asset Bundles.