Infrastructure • OpenAI Agents SDK Official

Security for
OpenAI
Agents

Preventing unauthorized tool calls and PII exfiltration in autonomous OpenAI agent loops.

Quick Install

pip install suprawall
GitHub Documentation

Implementation

from openai import OpenAI
from suprawall.openai import SupraWallProtector

# Initialize the protector
sw = SupraWallProtector(api_key="sw_live_...")

# Wrap your OpenAI Assistant or Agent
# This intercepts every tool execution at the SDK level
protected_client = sw.protect(OpenAI())

# Agent tool calls are now policy-governed
protected_client.beta.assistants.create(
    name="Financial Analyst",
    tools=[{"type": "function", "function": ...}],
    model="gpt-4-turbo"
)

# Intercept loops and block unauthorized tool actions
# No prompt-based security needed. It's deterministic.

Deterministic Interception

Native support for OpenAI Assistant tool calls. Block execution even if the model attempts to override prompt-based instructions.

Real-time Auditing

Automatic logging of every tool call, argument, and result. Satisfies Article 12 of the EU AI Act.

Human Override Queue

Establish human-in-the-loop approvals for sensitive operations like refund_payment or delete_user.

Cross-Agent Observability

Single pane of glass for all OpenAI agents across your entire team or swarm.

Secure Your
OpenAI Agents.

Start Building for Free