Multi-Agent Governance • CrewAI
Security for
CrewAI
Swarms
Protecting multi-agent tool execution from recursive loops and rogue actions.
Quick Install
pip install suprawall
Implementation
from crewai import Agent, Crew, Task
from suprawall.crewai import SupraWallProtector
# Initialize the protector
sw = SupraWallProtector(api_key="sw_live_...")
# Configure your agents normally
researcher = Agent(...)
analyst = Agent(...)
# Wrap the Crew to intercept all tool calls across all agents
crew = Crew(
agents=[researcher, analyst],
tasks=[task1, task2],
step_callback=sw.step_callback # Intercept each thought/action
)
# Deterministic security for the entire swarm
crew.kickoff()Recursive Guard
Prevent agents from getting stuck in infinite tool-call loops that drain budget.
Tool Permissions
Restrict specific tools like 'Terminal' or 'FileWrite' to specific agents in the crew.
Swarm Audit
Unified audit trail for all agents within a crew, showing the complete chain of thought.