CrewAI TrustGuard Integration

Protect your CrewAI agents from prompt injection, malicious web content, and other AI security threats.

Installation

pip install crewai-trustguard

Quick Start

1. Add Security Tools to Your Agents

from crewai import Agent, Task, Crew
from crewai_trustguard import TrustGuardURLTool, TrustGuardDocumentTool

# Create protected tools
url_scanner = TrustGuardURLTool(api_key="ta_xxx...")
doc_scanner = TrustGuardDocumentTool(api_key="ta_xxx...")

# Add to your agents
researcher = Agent(
    role="Research Analyst",
    goal="Research topics safely from the web",
    backstory="An expert researcher who always verifies content safety.",
    tools=[url_scanner, doc_scanner],
    verbose=True
)

# Your agent will now scan web content for threats before processing

2. Use Convenience Factory Functions

from crewai_trustguard import create_protected_scraper

# One-liner to create a secure web scraper
scraper = create_protected_scraper(api_key="ta_xxx...")

researcher = Agent(
    role="Researcher",
    tools=[scraper],
    ...
)

3. Use Middleware for Custom Workflows

from crewai_trustguard import TrustGuardMiddleware, ThreatDetectedError

middleware = TrustGuardMiddleware(api_key="ta_xxx...")

# Scan any content
result = middleware.scan(web_content, source_type="web")
if result.is_safe:
    process(web_content)
else:
    print(f"Threat detected: {result.threats}")

# Or scan and raise on threat
try:
    safe_content = middleware.scan_or_raise(document_text, source_type="document")
    process(safe_content)
except ThreatDetectedError as e:
    print(f"Blocked: {e.reasoning}")

Available Tools

Configuration Options

Tool Options

url_tool = TrustGuardURLTool(
    api_key="ta_xxx...",          # Your TrustGuard API key
    timeout=30.0,                  # Request timeout in seconds
    on_threat="report",            # "report", "raise", or "sanitize"
)

Middleware Options

middleware = TrustGuardMiddleware(
    api_key="ta_xxx...",
    strict_mode=False,             # True = block on MEDIUM threats
    on_threat="log",               # "log", "raise", or "silent"
    on_scan_error="warn",          # "warn", "raise", or "allow"
)

Example: Protected Research Crew

from crewai import Agent, Task, Crew
from crewai_trustguard import (
    TrustGuardURLTool,
    TrustGuardDocumentTool,
    TrustGuardRAGTool,
)

# Set up protected tools
url_tool = TrustGuardURLTool(api_key="ta_xxx...")
doc_tool = TrustGuardDocumentTool(api_key="ta_xxx...")
rag_tool = TrustGuardRAGTool(api_key="ta_xxx...")

# Create agents with security tools
researcher = Agent(
    role="Security-Aware Researcher",
    goal="Research topics while protecting against malicious content",
    backstory="A thorough researcher who always scans content for threats.",
    tools=[url_tool],
    verbose=True
)

analyst = Agent(
    role="Document Analyst",
    goal="Analyze documents safely",
    backstory="An analyst who verifies document safety before processing.",
    tools=[doc_tool, rag_tool],
    verbose=True
)

# Create tasks
research_task = Task(
    description="Research the latest AI security threats from trusted sources.",
    expected_output="A summary of current AI security threats.",
    agent=researcher
)

analysis_task = Task(
    description="Analyze the research findings and identify key patterns.",
    expected_output="Key patterns and recommendations.",
    agent=analyst
)

# Run the crew
crew = Crew(
    agents=[researcher, analyst],
    tasks=[research_task, analysis_task],
    verbose=True
)

result = crew.kickoff()

Wrapping Custom Functions

You can also wrap existing functions to add automatic scanning:

from crewai_trustguard import TrustGuardMiddleware
import requests

middleware = TrustGuardMiddleware(api_key="ta_xxx...")

# Original function
def fetch_webpage(url):
    return requests.get(url).text

# Wrap it for automatic scanning
protected_fetch = middleware.wrap_function(
    fetch_webpage,
    source_type="web",
    scan_output=True
)

# Now all fetches are automatically scanned
content = protected_fetch("https://example.com")  # Auto-scanned!

Threat Types Detected

TrustGuard detects multiple threat categories:

• Prompt Injection: Hidden instructions attempting to manipulate agents
• Jailbreak Attempts: Attempts to bypass agent restrictions
• Data Exfiltration: Patterns designed to leak sensitive data
• Memory Poisoning: Malicious content targeting agent memory
• RAG Poisoning: Document content designed to corrupt vector stores
• Tool Description Poisoning: Malicious tool descriptions
• Identity Manipulation: Attempts to override agent identity

API Reference

See TrustAgents Documentation for full API reference.

Support

• Documentation: https://trustagents.dev/docs
• Issues: https://github.com/trustagents/crewai-trustguard/issues
• Discord: https://discord.gg/trustagents

License

MIT License - see LICENSE file for details.