Support Service Account Authentication for Vertex AI Without gcloud CLI

[bakeryproducts]

Currently, it appears that using Vertex AI requires the gcloud CLI. url
What is the recommended way to use the ADK with Vertex AI models programmatically, without relying on the CLI tool?
for example, from a remote machine running in Docker outside of GCP?

My current approach (outside of ADK) is to create a Google service account and use the following Python code:

credentials = Credentials.from_service_account_file(GOOGLE_APPLICATION_CREDENTIALS)
vertexai.init(project=project, location=location, credentials=credentials)

Is it possible to support authentication with a Google service account for Vertex AI, so that the CLI is not required?

[boyangsvl]

You may use the GOOGLE_APPLICATION_CREDENTIALS environment variable. See https://cloud.google.com/docs/authentication/application-default-credentials. Moving this to discussions.

[bakeryproducts]

it works with GOOGLE_APPLICATION_CREDENTIALS var. Thank you!

ps
I end up with this .env:

GOOGLE_GENAI_USE_VERTEXAI=True
GOOGLE_APPLICATION_CREDENTIALS=sa_creds.json
GOOGLE_CLOUD_PROJECT=name
GOOGLE_CLOUD_LOCATION=us-central1

sa_creds.json is a google service account creds, one can find instructions how to create SA in google access control / IAM docs

[rick-bm]

@boyangsvl Our security team does not allow placing the credentials.json file inside the Docker pod. Is there a way to achieve the same effect by passing the content of the credentials.json through a configuration center? For example, the content of credentials.json can be delivered via the configuration center and then assembled locally to pass it to the ADK initialization.