Allow implicit or default credentials for GCP resources (Vertex AI, GCS etc). As far as the application is concerned, this credential type is one that has no service key. It should just work, the same way as when you run gcloud in terminal.
It simplifies configuration for these cases:
Deployment on local computer if I am already logged in to GCP
Where credentials are supplied via the GOOGLE_APPLICATION_CREDENTIALS environment variable that gcloud libraries look for
Google Kubernetes deployments that use Workload Id as the recommended way to set up access to GCP resources in GKE. It enables the gcloud libs to authenticate transparently against the cluster's metadata server, the roles are bound directly to the serviceAccount the pods run under.
This could be passed down to LiteLLM config as vertex_credentials: "os.environ/GOOGLE_APPLICATION_CREDENTIALS"
Allow implicit or default credentials for GCP resources (Vertex AI, GCS etc). As far as the application is concerned, this credential type is one that has no service key. It should just work, the same way as when you run
gcloud
in terminal.It simplifies configuration for these cases:
This could be passed down to LiteLLM config as
vertex_credentials: "os.environ/GOOGLE_APPLICATION_CREDENTIALS"