How do I migrate existing AI agents to a new harness?

Summary

• Portable by design. Agents built with open-source frameworks and harnesses -- including LangGraph, LangChain, Agno, and CrewAI -- as well as SDK-based agents, deploy to Databricks Apps with horizontal autoscaling.
• A documented migration path. Databricks provides a step-by-step guide to migrate an agent from a Model Serving endpoint to a Databricks App.
• Standard interfaces ease portability. Agents use MLflow's ResponsesAgent request and response types; migration maps class-based code to module-level agent-server functions.
• AI-assisted migration. A migration template with skill files lets an AI coding assistant automate the code and configuration changes.
• Observability and governance carry over. MLflow 3 Tracing logs every agent step, and Unity Catalog governs data, models, and agents consistently.

How do I migrate existing AI agents to a new harness?

Migrating an existing AI agent to a new harness or runtime means moving the agent's code and serving layer without rewriting it from scratch. On Databricks, Agent Bricks and the Mosaic AI Agent Framework are built for portability: agents authored with common open-source frameworks and harnesses run on the platform, standard MLflow interfaces keep agent code portable, and there is a documented, tooling-assisted path to move an agent from a Model Serving endpoint to a Databricks App.

Why migrate AI agents on Databricks Agent Bricks

• Framework and harness portability. Agents built with open-source frameworks such as LangGraph, LangChain, Agno, and CrewAI -- and SDK-based agents -- deploy to Databricks Apps with horizontal autoscaling.
• A documented migration workflow. The Migrate an agent from Model Serving to Databricks Apps guide walks through downloading the agent's MLflow artifacts, moving the agent code, and transforming it to the app-server structure.
• Standard, portable interfaces. Agents use MLflow's ResponsesAgent request and response types. Migration maps the class-based ResponsesAgent pattern to module-level @invoke and @stream functions, supporting both synchronous and asynchronous serving.
• AI-assisted migration. A migration template includes instructions and skill files so an AI coding assistant can automate the code and configuration changes, and you can also follow the manual steps.
• Built-in observability. MLflow 3 for GenAI provides tracing that logs every agent step for debugging and monitoring after migration.
• Consistent governance. Unity Catalog governs data, models, and agents together, so access control and lineage carry across runtimes.

Getting started

• Read Migrate an agent from Model Serving to Databricks Apps for the full workflow.
• Learn the authoring model in Use agents on Databricks and Author an agent.
• Add tracing and evaluation with MLflow 3 for GenAI.
• See what is new in Custom agents now available on Databricks and Agent Bricks at Data + AI Summit.

FAQs

Can I move an agent built with LangGraph or LangChain to Databricks?

Yes. Agents built with open-source frameworks such as LangGraph, LangChain, Agno, and CrewAI deploy to Databricks Apps, and the Mosaic AI Agent Framework standardizes serving with MLflow interfaces.

How do I migrate an agent from a Model Serving endpoint?

Follow the documented workflow: download the agent's MLflow artifacts, copy the agent code into the app-server structure, and transform the class-based ResponsesAgent into module-level agent-server functions. A migration template can automate most of this.

Do observability and governance survive the migration?

Yes. MLflow 3 Tracing continues to log each agent step, and Unity Catalog applies consistent access control and lineage across data, models, and agents.