OpenAI Agents Integration

Overview

AgentTrust provides agent_guard, OpenAIAgentsGuard, and SwarmGuard for the OpenAI Agents SDK and Swarm multi-agent patterns.

Tier: Team+

Why It Matters

OpenAI's agent framework runs tool-use loops and multi-agent handoffs. Guards validate outputs at handoff boundaries.

Prerequisites

pip install agentrust-sdk
# Team+ API key
# openai-agents or swarm package

Step-by-Step Guide

from agentrust_sdk import agent_guard, OpenAIAgentsGuard, SwarmGuard

# Single agent
guard = OpenAIAgentsGuard(agent_id="openai-agent")

# Swarm multi-agent
swarm_guard = SwarmGuard(agent_id="swarm-orchestrator")

# Decorator-style
@agent_guard(agent_id="my-openai-agent")
def run_agent(user, input):
    return agent.run(input)

Examples

For basic OpenAI SDK (non-Agents), use auto_instrument() — see OpenAI Direct example.

Best Practices

• Use SwarmGuard for multi-agent handoff validation
• Distinguish OpenAI SDK (auto_instrument) from OpenAI Agents SDK (this adapter)
• Set per-agent IDs in swarm configurations

Common Mistakes

• Using auto_instrument() for OpenAI Agents SDK (different package)
• Missing Team API key

Troubleshooting

Related Resources

• Examples: OpenAI Direct
• Auto-Instrumentation