OpenAI Agents SDK Guardrails

Brane can be used as a runtime policy layer for OpenAI Agents SDK function tools by wrapping tool functions as Brane capabilities. The Brane policy runs before the tool function executes and can allow or deny the attempted action. This gives OpenAI Agents SDK applications a policy-as-code layer for production tool use.

How Brane Fits

The OpenAI Agents SDK helps define and run agents. Brane governs what those agents are allowed to do when they attempt actions.

OpenAI agent chooses function tool -> Brane intercepts tool function -> policy decides -> tool runs or is blocked

Example Pattern

from brane import Decision, Runtime

runtime = Runtime(agent_id="openai-support-agent", environment="prod")

@runtime.capability(name="refund_customer", type="tool", risk="high")
def refund_customer(customer_id: str, amount_usd: float):
    return {"refunded": True, "amount": amount_usd}

@runtime.before_capability("refund_customer")
def refund_limit(ctx):
    if ctx.arg("amount_usd", 0) > 100:
        return Decision(type="deny", reason="Refund exceeds $100 limit")
    return Decision(type="allow")

Expose refund_customer as the function tool in your agent framework. Brane remains responsible for runtime policy enforcement.

What This Protects

Destructive tools
Financial actions
Cross-tenant access
Production-only workflows
Database queries
External API calls
Actions requiring approval

Current Status

The framework-independent Python runtime works today. A dedicated OpenAI Agents SDK adapter is planned.

Start Here

Core Model

Answer Guides

Build With Python

Recipes

Reference

OpenAI Agents SDK Guardrails

How Brane Fits

Example Pattern

What This Protects

Current Status

​How Brane Fits

​Example Pattern

​What This Protects

​Current Status

​Related Pages

How Brane Fits

Example Pattern

What This Protects

Current Status

Related Pages