OpenAI Agents SDK

If you build agents with the OpenAI Agents SDK, attach Kaizen with a run hook. Every tool the agent calls flows to Kaizen, so the Observer learns its behavior and flags deviations.

pip install "kaizen-security[openai-agents]"

from agents import Agent, Runner
from kaizen_security import Kaizen
from kaizen_security.integrations.openai_agents import KaizenHooks

kz = Kaizen(api_key="kz_live_...", agent="support-bot")

agent = Agent(name="support-bot", instructions="...", tools=[...])

await Runner.run(agent, "handle this ticket", hooks=KaizenHooks(kz))

Observe vs enforce

By default the hook observes: it inspects and reports every tool call without changing behavior.

KaizenHooks(kz, enforce=True)

With enforce=True, a tool the policy blocks is stopped before it runs (via the SDK's tool-rejection path).