Configuration
The DIAL Vertex AI Adapter is configured with environment variables on its container, while each model is configured in DIAL Core. This reference lists every variable: Google Cloud credentials, Claude defaults, retries, and the upstream connection pool. It is for platform engineers who run the adapter; per-upstream overrides are in Regions, projects, and load balancing.
Google Cloud
| Variable | Default | Description |
|---|---|---|
GOOGLE_APPLICATION_CREDENTIALS | — | The path to a JSON file with credentials, such as a service account key. |
GCP_PROJECT_ID | — | The Google Cloud project. |
GCP_DEFAULT_REGION | — | The default Vertex AI region, for example us-central1. |
DEFAULT_REGION | — | Deprecated. Use GCP_DEFAULT_REGION; ignored when that is set. |
General
| Variable | Default | Description |
|---|---|---|
DIAL_URL | — | The URL of DIAL Core. Optional; used to read images stored in DIAL file storage. |
LOG_LEVEL | INFO | The log level. Use DEBUG for development only. |
CLAUDE_DEFAULT_MAX_TOKENS | 1536 | The default max_tokens for Claude. Prefer a per-model default in DIAL Core instead. |
GOOGLE_GENAI_MAX_RETRY_ATTEMPTS | 0 | Retries of Google GenAI requests on a retriable error. |
ANTHROPIC_MAX_RETRY_ATTEMPTS | 0 | Retries of Anthropic requests on a retriable error. |
COMPATIBILITY_MAPPING | {} | Deprecated. A JSON map from unsupported to supported model IDs. Use compatible_model_id in DIAL Core instead; see Compatibility mode. |
Server and connections
| Variable | Default | Description |
|---|---|---|
WEB_CONCURRENCY | 1 | The number of server workers. |
TIMEOUT_KEEP_ALIVE | 70 | How long an idle keep-alive connection stays open, in seconds. Keep it 10 seconds above DIAL Core's client.keepAliveTimeout, 60 by default, so Core always closes first. |
HTTP_MAX_CONNECTIONS | 250 | The most concurrent upstream connections. The Anthropic, Mistral, and Google GenAI clients share one pool, so this caps the adapter's total upstream concurrency. |
HTTP_MAX_KEEPALIVE_CONNECTIONS | 75 | The most idle connections kept for reuse. Ignored with HTTP_TRANSPORT=aiohttp. |
HTTP_POOL_TIMEOUT | 10 | How long a request waits for a free connection, in seconds. On timeout, the adapter answers 503. |
HTTP_TRANSPORT | aiohttp | The shared connection pool: aiohttp or httpx. httpx costs more CPU with many concurrent streams. |
Logging
Logging comes from the DIAL SDK: the format, structured JSON output, trace correlation, and OpenTelemetry log export are set by the DIAL_SDK_* variables in the DIAL SDK logging documentation. To log the Anthropic SDK, set ANTHROPIC_LOG=debug.
Example
adapter-dial-vertexai:
image: epam/ai-dial-adapter-vertexai:0.41.2
environment:
DIAL_URL: http://core:8080
GCP_PROJECT_ID: <GCP_PROJECT_ID>
GCP_DEFAULT_REGION: us-central1
GOOGLE_APPLICATION_CREDENTIALS: /opt/credentials/google.json
GOOGLE_GENAI_MAX_RETRY_ATTEMPTS: "2"
volumes:
- ./credentials:/opt/credentials:ro
Further reading
- DIAL Vertex AI Adapter capabilities — regions, authentication, and compatibility mode
- Vertex AI model deployment — prepare Google Cloud for DIAL
- DIAL Vertex AI Adapter supported models — the model IDs to use