Vertex AI dual-path

Configure Gemini on Vertex (project/location publisher URLs) vs AI Studio, with service account and WIF auth.

Run native Gemini (kind: google) against Vertex AI or AI Studio by choosing the right base_url and auth mode.

What it is

Path	Host	Auth
AI Studio	`generativelanguage.googleapis.com/v1beta`	API key or OAuth
Vertex	`{LOCATION}-aiplatform.googleapis.com/v1/projects/{P}/locations/{L}/publishers/google`	ADC / service account / token file

Client paths stay the same (/v1beta/models/{model}:generateContent, Live, embeddings translate, platform proxies). Only the provider base_url + auth change.

How it works

Egress builds relative paths like /models/{id}:generateContent. That is appended to base_url, so Vertex base_url must already include the publisher prefix (no extra /v1beta segment unless your endpoint requires it).

Helper (Go library / tests):

base := config.VertexBaseURL("my-project", "us-central1")
// https://us-central1-aiplatform.googleapis.com/v1/projects/my-project/locations/us-central1/publishers/google

config.VertexBaseURL("my-project", "global")
// https://aiplatform.googleapis.com/v1/projects/my-project/locations/global/publishers/google

Configuration

AI Studio (API key)

providers:
  google:
    kind: google
    base_url: "https://generativelanguage.googleapis.com/v1beta"
    api_key_env: GEMINI_API_KEY
defaults:
  google_dialect: google

Vertex + service account JSON

providers:
  vertex:
    kind: google
    base_url: "https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT/locations/us-central1/publishers/google"
    auth: service_account
    service_account_file: /secrets/vertex-sa.json
defaults:
  google_dialect: vertex

Mount the SA file read-only. The gateway signs a JWT and exchanges it for an access token (no Cloud SDK).

Vertex + WIF / token file

providers:
  vertex:
    kind: google
    base_url: "https://us-central1-aiplatform.googleapis.com/v1/projects/PROJECT/locations/us-central1/publishers/google"
    auth: adc
    token_file: /var/run/secrets/gcp-access-token

See WIF & workload identity.

How to call

Same as AI Studio from the client:

curl -sS "$GW/v1beta/models/gemini-2.0-flash:generateContent" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $EDGE_OR_TOKEN" \
  -d '{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}'

With api_key_env / SA on the server, clients only need edge auth.

IAM notes

Prefer least privilege (roles/aiplatform.user or custom).
Prefer Workload Identity over long-lived JSON keys when possible.
Do not embed end-user OAuth for multi-tenant products without reviewing ToS.