Example — GPT-4o (OpenAI)
---
spec_version: "1.2"
model_id: "gpt-4o"
version: "1.0.0"
status: "active"
meta:
name: "GPT-4o (OpenAI)"
owner: "platform-ml-team"
last_updated: "2026-01-15"
provider:
type: "openai"
config:
model: "gpt-4o"
credentials:
source: "aws_secrets_manager"
secret_id: "prod/openai/api-key"
region: "us-east-1"
key: "api_key"
capabilities:
context_window: 128000
max_output_tokens: 16384
supports_tools: true
supports_system_prompt: true
supports_streaming: true
modalities: ["text", "image"]
defaults:
temperature: 0.7
max_tokens: 4096
---
# Description
GPT-4o is OpenAI's flagship multimodal model, accessed directly through the OpenAI API. It supports text and image inputs, tool use, and a 128,000-token context window.
Use this model when agents must interoperate with OpenAI-specific features, when your organization has an existing OpenAI contract, or when evaluating output quality across providers.
The credentials are fetched at runtime from the backend declared in `credentials.source`. They must never appear in any AML file. To use an environment variable during local development, set `source: "env"` and add `name: "OPENAI_API_KEY"`.
Notes
- Credential model: The runtime resolves
credentials at agent startup. The example above reads the API key from AWS Secrets Manager (prod/openai/api-key, key api_key). For development, substitute source: "env" and name: "OPENAI_API_KEY" to read from an environment variable instead. The key value must never appear in any AML file.
- Secret rotation: Rotate the secret in AWS Secrets Manager (or whichever backend is declared). No AML files need to change.
- Rate limits: OpenAI rate limits apply per organization and tier. For high-throughput agents, consider
litellm with load balancing across multiple keys.
- Alternative endpoint: To use an Azure OpenAI deployment instead, set
base_url to your Azure endpoint and update credentials accordingly. The model field should match your Azure deployment name.