genai-hub-foundation-models

Rules

1. Fetch the live model list when the model name is unknown or unconfirmed — skip if the user already named a specific model.
2. executable-id and model-name values from this skill are the inputs needed by aicore-lifecycle-management to create a deployment.
3. If credentials are not configured, invoke aicore-admin-resources first.
4. Before executing any curl example, get a Bearer token with:

   export TOKEN=$(uv run skills/aicore-admin-resources/scripts/get_token.py)
   

   Never add 2>&1 or pipe the command — that prints the raw JWT into the conversation.

---

List Available Models

# Show all options and examples:
uv run scripts/list_foundation_models.py --help

# One provider only (e.g. Azure OpenAI):
uv run scripts/list_foundation_models.py --executable-id azure-openai

# Detailed view (versions, capabilities, context length, streaming support):
uv run scripts/list_foundation_models.py --response-format detailed

# Machine-readable JSON for scripting:
uv run scripts/list_foundation_models.py --json | jq '.[].model'
uv run scripts/list_foundation_models.py --response-format detailed --json | jq '.[] | select(.model == "gpt-4o") | .versions'

# Models supported by the Orchestration service only:
uv run scripts/list_foundation_models.py --scenario-id orchestration

---

Provider → Executable ID Mapping

Provider	--executable-id
Azure OpenAI	azure-openai
Anthropic	aws-bedrock
Google	gcp-vertexai
Amazon Bedrock	aws-bedrock
Mistral AI	aicore-mistralai
Cohere	aicore-cohere
Perplexity	perplexity-ai
NVIDIA	aicore-nvidia
Open Source	aicore-opensource
SAP	aicore-sap

Supported models in your instance may differ from this table — use the script to get the live list.

---

Create a Deployment

Once you have the executable-id and model-name, check whether a RUNNING deployment for that model already exists:

uv run skills/aicore-lifecycle-management/scripts/get_deployments.py \
  --scenario foundation-models --status RUNNING --json

If one is found, use AskUserQuestion to ask whether to reuse it or create a new one. Then invoke aicore-lifecycle-management with both values.

The modelName value must exactly match the catalog output — a mismatch causes an "Invalid Configuration" error at deployment time, not at configuration creation.

---

When to Load Reference Files

Trigger	Load
Curl inference examples, provider-specific endpoint paths, or about to call a deployed model	references/MODELS.md

---

Finding the Deployment URL for a Curl Call

1. Get the executable-id — look it up in the Provider → Executable ID table above.

2. Find the running deployment:

   uv run skills/aicore-lifecycle-management/scripts/get_deployments.py \
     --scenario foundation-models --executable-id <executable-id> --status RUNNING --json
   # use .deployment_url and .resource_group from the matching entry
   

3. See references/MODELS.md for export setup and curl examples per provider.

---

Handoffs

• Credential/auth error? → invoke aicore-admin-resources
• Ready to create a deployment with the model you found? → invoke aicore-lifecycle-management