<working-directory>/
├── original_mlflow_model/    # Downloaded artifacts from Model Serving
│   ├── MLmodel
│   ├── code/
│   │   └── agent.py
│   ├── input_example.json
│   └── requirements.txt
│
└── <app-name>/               # New Databricks App (ready to deploy)
    ├── agent_server/
    │   ├── agent.py          # Migrated agent code
    │   └── ...
    ├── app.yaml
    ├── databricks.yml        # Bundle config with resources
    ├── pyproject.toml
    ├── requirements.txt
    └── ...

databricks current-user me --profile <profile>

databricks auth login --profile <profile>

undefined

Look for `served_entities[0].entity_name` (model name) and `entity_version` in the response. Find the entity with 100% traffic in `traffic_config.routes`.

DATABRICKS_CONFIG_PROFILE=<profile> uv run --no-project \
  --with "mlflow[databricks]>=2.15.0" \
  --with "databricks-sdk>=0.30.0" \
  python3 << 'EOF'
import mlflow

mlflow.set_tracking_uri("databricks")

undefined

undefined

**Examine the `/code` folder** - contains all code dependencies logged via `code_paths=["..."]`:

```bash

**Examine the `/artifacts` folder** (if present) - contains artifacts logged via `artifacts={...}`:

```bash

> **Important:** Take note of ALL files in `/code` and `/artifacts`. You will need to copy these to the migrated app and ensure imports still work correctly.

./original_mlflow_model/
├── MLmodel              # Model metadata and resource requirements
├── code/                # Code logged via code_paths=["..."]
│   ├── agent.py         # Main agent implementation
│   ├── utils.py         # (optional) Helper modules
│   ├── tools.py         # (optional) Custom tool definitions
│   └── ...              # Any other code dependencies
├── artifacts/           # (optional) Artifacts logged via artifacts={...}
│   ├── config.yaml      # (optional) Configuration files
│   ├── prompts/         # (optional) Prompt templates
│   └── ...              # Any other artifacts (data files, etc.)
├── input_example.json   # Sample request for testing
├── requirements.txt     # Original dependencies
└── ...

databricks auth login --profile <profile>

undefined

---

from mlflow.pyfunc import ResponsesAgent, ResponsesAgentRequest, ResponsesAgentResponse

class MyAgent(ResponsesAgent):
    def predict(self, request: ResponsesAgentRequest, params=None) -> ResponsesAgentResponse:
        # Synchronous implementation
        ...
        return ResponsesAgentResponse(output=outputs)

    def predict_stream(self, request: ResponsesAgentRequest, params=None):
        # Synchronous generator
        for chunk in ...:
            yield ResponsesAgentStreamEvent(...)

from mlflow.genai.agent_server import invoke, stream
from mlflow.types.responses import (
    ResponsesAgentRequest,
    ResponsesAgentResponse,
    ResponsesAgentStreamEvent,
)

@invoke()
async def non_streaming(request: ResponsesAgentRequest) -> ResponsesAgentResponse:
    # Async implementation - typically calls streaming() and collects results
    outputs = [
        event.item
        async for event in streaming(request)
        if event.type == "response.output_item.done"
    ]
    return ResponsesAgentResponse(output=outputs)

@stream()
async def streaming(request: ResponsesAgentRequest) -> AsyncGenerator[ResponsesAgentStreamEvent, None]:
    # Async generator
    async for event in ...:
        yield event

from mlflow.genai.agent_server import invoke, stream
from mlflow.types.responses import (
    ResponsesAgentRequest,
    ResponsesAgentResponse,
    ResponsesAgentStreamEvent,
)

@invoke()
def non_streaming(request: ResponsesAgentRequest) -> ResponsesAgentResponse:
    # Same sync logic from original predict(), extracted from the class
    ...
    return ResponsesAgentResponse(output=outputs)

@stream()
def streaming(request: ResponsesAgentRequest):
    # Same sync generator from original predict_stream(), extracted from the class
    for chunk in ...:
        yield ResponsesAgentStreamEvent(...)

undefined

---

undefined

**Copy artifacts (if present):**

```bash

**Fix import paths after copying:**

When code files are moved, imports may break. Check and update imports in all copied files:

```python

> **Important:** Review each copied file and ensure all imports resolve correctly. The most common issues are:
> - Relative imports that assumed a different directory structure
> - Hardcoded file paths to artifacts
> - Missing `__init__.py` files for package imports

from mlflow.genai.agent_server import invoke, stream
from mlflow.types.responses import (
    ResponsesAgentRequest,
    ResponsesAgentResponse,
    ResponsesAgentStreamEvent,
)

2. **Key changes from class to functions:**
   - Remove the `class MyAgent(ResponsesAgent):` wrapper
   - Remove `self` parameter from all methods
   - Move `__init__` logic (client creation, tool setup) to module-level code
   - Replace `self.some_attribute` with module-level variables
   - Add `@invoke()` decorator to the non-streaming function
   - Add `@stream()` decorator to the streaming function

3. **Keep all other code as-is** — no need to convert sync calls to async, no need to change `for` to `async for`, no need to add `await`.

---

undefined

undefined

cd <app-name>
uv sync

echo "uv" > requirements.txt

uv run quickstart

undefined

To manually create an MLflow experiment:

```bash
databricks experiments create-experiment "/Users/<your-username>/<app-name>" --profile <profile>

cd <app-name>
uv run start-app

undefined

Example content:
```json
{"input": [{"role": "user", "content": "What is an LLM agent?"}], "custom_inputs": {"thread_id": "example-thread-123"}}

undefined

undefined

undefined

undefined

undefined

undefined

resources:
  api_version: '1'
  databricks:
    lakebase:
    - name: lakebase
    serving_endpoint:
    - name: databricks-claude-sonnet-4-5

resources:
  experiments:
    agent_migration_experiment:
      name: /Users/${workspace.current_user.userName}/${bundle.name}-${bundle.target}

  apps:
    agent_migration:
      name: "<app-name>"  # Update to user's app name
      description: "Migrated agent from Model Serving to Databricks Apps"
      source_code_path: ./
      resources:
        - name: 'experiment'
          experiment:
            experiment_id: "${resources.experiments.agent_migration_experiment.id}"
            permission: 'CAN_MANAGE'
        - name: 'serving-endpoint'
          serving_endpoint:
            name: 'databricks-claude-sonnet-4-5'
            permission: 'CAN_QUERY'
        - name: 'python-exec'
          uc_securable:
            securable_full_name: 'system.ai.python_exec'
            securable_type: 'FUNCTION'
            permission: 'EXECUTE'

targets:
  prod:
    resources:
      apps:
        agent_migration:
          name: "<app-name>"  # Same name for production

        - name: 'database'
          database:
            database_name: 'databricks_postgres'
            instance_name: 'lakebase'
            permission: 'CAN_CONNECT_AND_CREATE'

undefined

> **Important:** `bundle deploy` only uploads files and configures resources. `bundle run` is **required** to actually start/restart the app with the new code. If you only run `deploy`, the app will continue running old code!

undefined

Once the deployed app responds successfully:

> **Task:** Mark "Test deployed app" as `completed`. Migration complete!

undefined

**"App already exists" error:**
If `databricks bundle deploy` fails because the app already exists, refer to the **deploy** skill for instructions on binding an existing app to the bundle.

---

<app-name>/
├── agent_server/
│   ├── __init__.py
│   ├── agent.py          # Main agent logic - THIS IS WHERE YOU MIGRATE TO
│   ├── start_server.py   # FastAPI server setup
│   ├── utils.py          # Helper utilities
│   └── evaluate_agent.py # Agent evaluation
├── scripts/
│   ├── __init__.py
│   ├── quickstart.py     # Setup script
│   └── start_app.py      # App startup
├── app.yaml              # Databricks Apps configuration
├── databricks.yml        # Databricks Asset Bundle configuration (resources, targets)
├── pyproject.toml        # Dependencies (for local dev with uv)
├── requirements.txt      # REQUIRED: Must contain "uv" for Databricks Apps
├── .env.example          # Environment template
└── README.md

class ChatAgent(ResponsesAgent):
    def predict(self, request, params=None):
        messages = to_chat_completions_input(request.input)
        response = self.llm.invoke(messages)
        return ResponsesAgentResponse(output=[...])

llm = ...  # Move class-level init to module level

@invoke()
def non_streaming(request: ResponsesAgentRequest) -> ResponsesAgentResponse:
    messages = to_chat_completions_input(request.input)
    response = llm.invoke(messages)
    return ResponsesAgentResponse(output=[...])

@stream()
def streaming(request: ResponsesAgentRequest):
    # Original predict_stream() body, with self. removed
    ...

@invoke()
async def non_streaming(request: ResponsesAgentRequest) -> ResponsesAgentResponse:
    outputs = [e.item async for e in streaming(request) if e.type == "response.output_item.done"]
    return ResponsesAgentResponse(output=outputs)

@stream()
async def streaming(request: ResponsesAgentRequest) -> AsyncGenerator[ResponsesAgentStreamEvent, None]:
    messages = {"messages": to_chat_completions_input([i.model_dump() for i in request.input])}
    agent = await init_agent()
    async for event in process_agent_astream_events(agent.astream(messages, stream_mode=["updates", "messages"])):
        yield event

from langchain_core.tools import tool

@tool
async def search_docs(query: str) -> str:
    """Search the documentation."""
    results = await vector_store.asimilarity_search(query)
    return format_results(results)

from langchain.agents import create_agent
from databricks_langchain import ChatDatabricks

async def init_agent():
    tools = await mcp_client.get_tools()  # MCP tools are async
    model = ChatDatabricks(endpoint=LLM_ENDPOINT_NAME)
    return create_agent(model=model, tools=tools, system_prompt=SYSTEM_PROMPT)

uv sync  # Reinstall dependencies

databricks auth login  # Re-authenticate