# Required
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_CLOUD_LOCATION=global

# Gateway auth: comma-separated API keys your apps use when calling this gateway.
# Example request header: Authorization: Bearer dev-local-key
GATEWAY_API_KEYS=dev-local-key
GATEWAY_ALLOW_UNAUTHENTICATED=false

# Model behavior: services must send real Gemini model IDs or configured aliases.
MODEL_CATALOG_PATH=config/models.json
MODEL_CATALOG_REFRESH_ON_START=true
ALLOW_ANY_GEMINI_MODEL=false

# Optional aliases if you want them. Keep empty if you want strict real model names only.
# MODEL_ALIASES=pro=gemini-3.1-pro-preview,tools=gemini-3.1-pro-preview-customtools
MODEL_ALIASES=

# Vertex endpoint base. Keep default unless you know you need a regional endpoint.
VERTEX_BASE_URL=https://aiplatform.googleapis.com

# Auth token source order:
# 1) VERTEX_ACCESS_TOKEN if set
# 2) GOOGLE_APPLICATION_CREDENTIALS service account JSON
# 3) Cloud Run/GCE metadata server
# 4) gcloud auth application-default print-access-token
VERTEX_ACCESS_TOKEN=
GOOGLE_APPLICATION_CREDENTIALS=secrets/llm-gateway-sa.json

# Server
PORT=8080
LOG_PATH=logs/requests.jsonl
REQUEST_TIMEOUT_SECONDS=180

# Lightweight Vertex retry policy for transient transport/upstream failures.
# Set max attempts to 1 to disable gateway retries.
VERTEX_RETRY_MAX_ATTEMPTS=3
VERTEX_RETRY_INITIAL_MS=250
VERTEX_RETRY_MAX_MS=2000