Inspecting build and deployment artifacts in Agent Engine

Google Cloud Build with AI Agents
, ,
1

I’m deploying an ADK agent to Agent Engine from source files as described here:

remote_agent = client.agent_engines.create(
    config=dict(
        {
            "source_packages": ["sputnik"],
            # ... omitted for brevity ...
        }
)

I believe this method zips up the listed packages (source code directories) into an archive and sends it off to the cloud for building.

The logs in Logs Explorer show that the archive is uploaded to Cloud Storage, and then it looks like aiplatform.googleapis.com%2Freasoning_engine_build runs a Cloud Build job which produces a container image.

All this seems to happen in Google-managed projects because I don’t see any Cloud Build jobs, Docker container repos, or new Cloud Storage buckets in my project. I tried downloading the source code archive using the Cloud Storage link from the logs but it returned the “unauthorised” error, so I don’t have access to that, either.

I would like to:

  • Retain a copy of the source code archive file before it is uploaded.
  • Download the Docker container image that the build produces.

How can I do that, please?

Thank you.

2

Hey,

Hope you’re keeping well.

When you deploy an ADK agent to Agent Engine, the build and containerization steps run entirely in Google-managed projects, so you won’t have direct access to the intermediate Cloud Storage objects or Cloud Build jobs. To retain your source archive, you’ll need to create your own zip or tarball locally before calling agent_engines.create, and store it in a bucket in your project. The container images produced by the managed build are stored in Google-managed Artifact Registry instances and aren’t exposed for direct download.

Thanks and regards,
Taz

3

Hi,

Could you, please, provide an example of how this API method can be invoked with an existing source archive zip or a tarball?

Thank you.

1 Like
4

Hi! When deploying ADK agents via Agent Engine, the source archive and resulting container image are managed in Google-owned projects, so direct access isn’t provided. To retain a copy of the source, create a local archive before deployment. For the container image, consider building it locally using the same source and configuration, which gives you full control over the image and storage.

5

If I build the local archive before deployment myself, how do I submit an existing local archive file to Agent Engine API for deployment, please?

If I build the container image locally, how then I deploy the image to Agent Engine, please?