AI models

Every cloud serves a different set of models, and some need a one-time customer step before they answer. Call getAvailableModels() at runtime instead of hardcoding an id.

The model ids below are Alien's own. The gateway maps each to the cloud's model id.

The native protocol decides the fastest path. Every model also answers on the other endpoints through translation. See Protocols.

AWS: Amazon BedrockCopy Anchor Link

45 models.

PrerequisitesCopy Anchor Link

Claude needs a model agreement per Claude model and an Anthropic use-case form submitted once for the account. Submitting the form from an AWS Organizations management account covers every member account. Neither step needs the console:

# once per account
aws bedrock put-use-case-for-model-access --form-data "$(base64 < form.json)"

# once per Claude model
OFFER=$(aws bedrock list-foundation-model-agreement-offers --model-id "$MODEL" \
  --query 'offers[0].offerToken' --output text)
aws bedrock create-foundation-model-agreement --offer-token "$OFFER" --model-id "$MODEL"

aws bedrock get-foundation-model-availability --model-id "$MODEL"

Agreements are usage-priced with no fixed fee, so accepting one costs nothing until a call is made. get-foundation-model-availability reports agreement, authorization, entitlement, and region status separately, which tells you which one is missing. Alien runs the same check in the background and shows the result in the resource's status.

The GPT-5 family runs on a separate Bedrock endpoint with its own quotas. Raising a Claude quota does nothing for GPT-5, and the reverse.

Check quota on a new AWS account. Quotas are per model and per region, and AWS states that new accounts might receive reduced quotas.

A fresh account can be far below the published defaults, and inference then fails in a way that looks like a broken integration. Check Service Quotas for Amazon Bedrock first.

Availability also varies by region. A model in us-east-1 may not exist in another region, so check Bedrock's regional availability table for the customer's region.

GCP: Vertex AICopy Anchor Link

15 models.

PrerequisitesCopy Anchor Link

To enable Claude, the customer opens Vertex AI Model Garden, finds the model, selects Enable, and accepts Anthropic's terms. It's per project, so a customer who did it for one project still has to do it for another.

Quota is per region, and the global endpoint has its own pool. A project with room in us-central1 can still be at zero on a global or multi-region endpoint. Check the quota page for the region the deployment runs in.

Gemini is GCP-only, though Google's Gemma models are on AWS.

Calls go to the location the deployment runs in. Google serves the 2.5 family in-region but the 3.x models only on its global location, so on a region-pinned deployment the 3.x models won't answer.

Azure: Azure AI FoundryCopy Anchor Link

13 models.

PrerequisitesCopy Anchor Link

Alien deploys gpt-4.1 and model-router. Every other model in the list answers once a deployment with that model's id as its name exists in the account.

Claude can't be automated: the first deployment requires accepting Marketplace terms, which is a portal action no API performs.

Quota is per subscription. It's tracked per model per subscription, and Global Standard deployments of the same model share one pool across regions. A new subscription can show 0 TPM in every region, which blocks deployment instead of throttling it.

Viewing quota needs the Cognitive Services Usages Reader role. Raising it needs Owner or Contributor.

model-router is one deployment that picks an underlying model per request, defaulting to the cheapest model within a narrow quality band of the best one. Its context window is that of the smallest model it can route to.

Check the region before deploying. Alien creates both deployments at the GlobalStandard tier and fails the resource if either fails. That happens mid-provision, after the account exists.

To check a region:

az rest --method get \
  --url "https://management.azure.com/subscriptions/<sub>/providers/Microsoft.CognitiveServices/locations/<region>/models?api-version=2024-10-01" \
  --query "value[?model.name=='model-router'] | [0].model.skus[?name=='GlobalStandard']"

Run it for gpt-4.1 too. An empty result means that region will fail.

External bindingsCopy Anchor Link

Kubernetes, Machines, and Local use an external binding with your key, so you can call whatever the key can reach. getAvailableModels() returns a short built-in list instead of querying the provider:

When a model is missingCopy Anchor Link

A model that fails with provider_access_unavailable, or shows as blocked in the resource's status, is usually one of these:

1. A one-time step is outstanding. The agreement and use-case form on AWS, Model Garden terms on GCP, a portal deployment on Azure. This is the usual answer for Claude.
2. Quota is zero. Common on new accounts and subscriptions, on all three clouds. On AWS a model without access reads 0 applied quota, so a quota increase does nothing until access exists.
3. The model isn't offered in that region. Availability differs per region on every cloud.

On AWS and GCP none of these fail the deploy. The other models keep working, and the model answers once the customer completes the step. On Azure, a failed Alien-created deployment fails the resource.

SourcesCopy Anchor Link

Cloud requirements change. These are the pages to recheck.

• Bedrock model access, automatic enablement
• Bedrock runtime quotas, regional availability by model
• Claude on Vertex AI
• Foundry quotas and limits, model router