Quickstart
1. Get an API key
Sign in at app.getkaizen.io, then open API keys and create one. It starts with kz_live_ and is shown once.
2. Install the SDK
pip install kaizen-security
3. Inspect an action
from kaizen_security import Kaizen
kz = Kaizen(api_key="kz_live_...", agent="support-bot")
verdict = kz.inspect(tool="export_file", publisher="external", target="45.9.148.108")
if verdict.blocked:
raise RuntimeError(verdict.reason)
Or wrap a tool so it is checked automatically:
@kz.guard(tool="send_email")
def send_email(to, body):
...
4. Watch verdicts
Back in the console, Verdicts shows every allow and block your agents produce, live.
5. Watch it catch an attack
To see Kaizen catch real attacks end to end, run the red-team corpus. It declares agents, runs normal activity, then runs ten classes of attack, and scores what Kaizen catches.
export KAIZEN_API_KEY=kz_live_...
python red-team/corpus.py
Detection scorecard: Kaizen caught 13/13 red-team actions (100%).
See how AI agents fail for the attack taxonomy.
What happens behind the scenes
Each inspect is evaluated against your policy locally (fast, fail-closed) and reported to Kaizen. The isolated Observer learns the agent's behavior over its first several actions, then flags anything that is a real departure from its norm.