Artifact reading after ADK deployment on vertex AI #1339

kn-engenuityai · 2025-06-12T06:43:22Z

kn-engenuityai
Jun 12, 2025

I want to save a CSV artifact using after_agent_callback and save it in a persistent storage.

Setup:

I created a Cloud Storage bucket for storing artifacts.
I passed this bucket to GcsArtifactService.
The approach works locally when I run the agent using a Runner.

Artifact-Saving Callback:

def after_agent_callback_3(*, callback_context: CallbackContext) -> None:
    # Extract state from callback context
    state = callback_context.state
    food_source = state.get("food_source")

    # Write CSV data to an in-memory buffer
    buf = io.StringIO()
    writer = csv.writer(buf)
    writer.writerow(["food_type", "region"])
    for food in food_source["sources"]:
        writer.writerow([food["food_type"], food["region"]])

    # Convert buffer to bytes and wrap in a Part with MIME type
    csv_bytes = buf.getvalue().encode('utf-8')
    csv_part = types.Part.from_bytes(
        data=csv_bytes,
        mime_type="text/csv",
    )

    # Save artifact using ADK's artifact service
    callback_context.save_artifact(
        filename="food.csv",
        artifact=csv_part
    )

    return types.ModelContent(
        parts=[
            types.Part(text='artifact saved'),
        ],
    )
    
root_agent = Agent(
    name="food_bot",
    model="gemini-2.0-flash",
    description="You are a .....",
    instruction=ROOT_AGENT_INSTRUCTION,
    output_schema=OutputSchema,
    after_agent_callback=after_agent_callback_3,
    output_key="food_source"
)

runner = Runner(
        agent=root_agent,
        app_name=app_name,
        session_service=session_service,
        artifact_service=GcsArtifactService(bucket_name=os.getenv("GOOGLE_CLOUD_ARTIFACT_BUCKET"))
    )

Deployment on Vertex AI
However, after deploying to Vertex AI using the following code, the artifact is not being saved, and I don’t see any object created in the Cloud Storage bucket. Although the code is working, states are created, can retrieve the states.


def artifact_service_builder():
    return GcsArtifactService(bucket_name=os.getenv("GOOGLE_CLOUD_ARTIFACT_BUCKET"))
    
app = reasoning_engines.AdkApp(
    agent=root_agent,
    enable_tracing=True,
    artifact_service_builder=artifact_service_builder  # Presumably returns GcsArtifactService
)

# Deploy to Agent Engine
remote_app = agent_engines.create(
    agent_engine=app,
    requirements=[
        "google-cloud-aiplatform[adk,agent_engines]",
        "google-cloud-storage",
    ],
    extra_packages=["./agents/abc"],
)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Artifact reading after ADK deployment on vertex AI #1339

Uh oh!

{{title}}

Uh oh!

Replies: 0 comments

Select a reply

Uh oh!

Artifact reading after ADK deployment on vertex AI #1339

Uh oh!

kn-engenuityai Jun 12, 2025

Replies: 0 comments

kn-engenuityai
Jun 12, 2025