interven-crewai 0.1.0

CrewAI integration for Interven AI firewall. Scan every tool call your CrewAI agents make — block malicious requests, redact PII/secrets, route risky actions to human approval.

interven-crewai

CrewAI integration for Interven — the AI firewall for agent tool calls.

Install

pip install interven-crewai

Quickstart

from crewai import Agent, Task, Crew
from crewai_tools import SerperDevTool

from interven_crewai import guard, InterventBlockedError

search = guard(SerperDevTool(), api_key="iv_live_...")

researcher = Agent(
    role="Security Researcher",
    goal="Find relevant AI security news",
    tools=[search],
    verbose=True,
)

task = Task(
    description="Summarize the latest prompt injection research",
    expected_output="A short summary",
    agent=researcher,
)

crew = Crew(agents=[researcher], tasks=[task])

try:
    result = crew.kickoff()
    print(result)
except InterventBlockedError as e:
    print(f"Crew blocked by Interven: {e}")

What happens per tool call

Interven decision	Behavior
ALLOW	tool runs unchanged
SANITIZE	tool runs with Interven's redacted arguments (SSN/secrets stripped)
DENY	raises InterventBlockedError
REQUIRE_APPROVAL	raises InterventBlockedError with approval URL

License

MIT