Google Vertex AI

Set up Vertex AI with a service-account key or application-default credentials, set project and region, and declare models.

Set up

Worker UI → AI models → Add → advanced kinds → vertex.
Credentials — one of: a service-account key (JSON) stored for the provider, GOOGLE_APPLICATION_CREDENTIALS, or gcloud auth application-default login on the worker machine.
extra.project (or taken from the credential) and extra.region (default us-central1; global uses aiplatform.googleapis.com).
Declare the models you use — Vertex has no simple per-project list of usable models.

The health check exchanges the credential for a Google OAuth token and reads the project's Vertex AI location.

How agents use it

Agents receive ANTHROPIC_VERTEX_PROJECT_ID, GOOGLE_CLOUD_PROJECT and CLOUD_ML_REGION. Claude Code runs with CLAUDE_CODE_USE_VERTEX=1; Gemini CLI uses Vertex directly as well.

Based on docs/providers/README.md in the core repository (v0.1.1).

Set up

Worker UI → AI models → Add → advanced kinds → vertex.
Credentials — one of: a service-account key (JSON) stored for the provider, GOOGLE_APPLICATION_CREDENTIALS, or gcloud auth application-default login on the worker machine.
extra.project (or taken from the credential) and extra.region (default us-central1; global uses aiplatform.googleapis.com).
Declare the models you use — Vertex has no simple per-project list of usable models.

The health check exchanges the credential for a Google OAuth token and reads the project's Vertex AI location.

How agents use it

Agents receive ANTHROPIC_VERTEX_PROJECT_ID, GOOGLE_CLOUD_PROJECT and CLOUD_ML_REGION. Claude Code runs with CLAUDE_CODE_USE_VERTEX=1; Gemini CLI uses Vertex directly as well.