Models

Agent Runtimes supports multiple AI model providers through pydantic-ai. Models are configured via environment variables and can be selected per-request (except for A2A protocol).

Supported Providers

Anthropic

Model ID	Display Name	Required Env Vars
anthropic:claude-sonnet-4-5	Claude Sonnet 4.5	ANTHROPIC_API_KEY
anthropic:claude-opus-4	Claude Opus 4	ANTHROPIC_API_KEY
anthropic:claude-sonnet-4-20250514	Claude Sonnet 4 (May 2025)	ANTHROPIC_API_KEY
anthropic:claude-3-5-haiku-20241022	Claude 3.5 Haiku	ANTHROPIC_API_KEY

OpenAI

Model ID	Display Name	Required Env Vars
openai:gpt-4o	GPT-4o	OPENAI_API_KEY
openai:gpt-4o-mini	GPT-4o Mini	OPENAI_API_KEY
openai:gpt-4-turbo	GPT-4 Turbo	OPENAI_API_KEY
openai:o1	o1	OPENAI_API_KEY
openai:o1-mini	o1 Mini	OPENAI_API_KEY
openai:o3-mini	o3 Mini	OPENAI_API_KEY

Azure OpenAI

Model ID	Display Name	Required Env Vars
azure:gpt-4o	GPT-4o (Azure)	AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT
azure:gpt-4o-mini	GPT-4o Mini (Azure)	AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT
azure:gpt-4-turbo	GPT-4 Turbo (Azure)	AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT
azure:o1	o1 (Azure)	AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT
azure:o1-mini	o1 Mini (Azure)	AZURE_OPENAI_API_KEY, AZURE_OPENAI_ENDPOINT

AWS Bedrock

Model ID	Display Name	Required Env Vars
bedrock:anthropic.claude-sonnet-4-5-20250514-v1:0	Claude Sonnet 4.5 (Bedrock)	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
bedrock:anthropic.claude-3-5-haiku-20241022-v1:0	Claude 3.5 Haiku (Bedrock)	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
bedrock:amazon.nova-pro-v1:0	Amazon Nova Pro	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
bedrock:amazon.nova-lite-v1:0	Amazon Nova Lite	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY
bedrock:meta.llama3-3-70b-instruct-v1:0	Llama 3.3 70B (Bedrock)	AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY

Configuration

Environment Variables

Set the required environment variables for your chosen provider:

# Anthropic
export ANTHROPIC_API_KEY="sk-ant-..."

# OpenAI
export OPENAI_API_KEY="sk-..."

# Azure OpenAI
export AZURE_OPENAI_API_KEY="..."
export AZURE_OPENAI_ENDPOINT="https://your-resource.openai.azure.com"
export AZURE_OPENAI_API_VERSION="2024-02-01"  # Optional

# AWS Bedrock
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."
export AWS_DEFAULT_REGION="us-east-1"  # Optional

Model Availability

Models are automatically marked as available or unavailable based on environment variables. The frontend displays:

• Available models: Selectable in the model dropdown
• Unavailable models: Grayed out with "Missing API key" message

Per-Request Model Selection

Most transports support changing the model on each request:

Transport	Per-Request Model Selection
AG-UI	✅ Supported
Vercel AI	✅ Supported
ACP	✅ Supported
A2A	❌ Not supported

How It Works

1. Frontend: User selects a model from the dropdown
2. Request: Model ID is sent with each message
3. Backend: Agent uses the specified model for that request

// Model is sent with each message
adapter.sendMessage(message, { model: 'anthropic:claude-sonnet-4-5' });

A2A Exception

The A2A protocol does not support per-request model selection. The model is fixed when the agent is created. See Transports > A2A Limitations for details.

Adding Custom Models

You can add custom model configurations by modifying the model list:

from agent_runtimes.types import AIModel

custom_model = AIModel(
    id="openai:gpt-4-custom",
    name="Custom GPT-4",
    builtin_tools=["search", "code_interpreter"],
    required_env_vars=["OPENAI_API_KEY"],
    is_available=True,
)

Model String Format

Models use the pydantic-ai format: provider:model_name

• anthropic:claude-sonnet-4-5 - Anthropic Claude
• openai:gpt-4o - OpenAI GPT-4o
• azure:gpt-4o - Azure OpenAI (deployment name)
• bedrock:anthropic.claude-sonnet-4-5-20250514-v1:0 - AWS Bedrock

Timeout Configuration

Models are created with configurable timeouts:

• Read/Write timeout: 60 seconds (default)
• Connect timeout: 30 seconds (Azure/OpenAI), 60 seconds (Anthropic)

from agent_runtimes.models import create_model_with_provider

model = create_model_with_provider(
    model_provider="azure-openai",
    model_name="gpt-4o",
    timeout=120.0,  # Custom timeout
)