agentguard-sdk 0.2.0

Python SDK for AI Agent Security Platform

AgentGuard Python SDK

Enterprise-grade security for AI agents - Runtime protection, policy enforcement, and comprehensive audit trails

🚀 Quick Start

pip install agentguard-sdk

from agentguard import AgentGuard

# Initialize the security client
guard = AgentGuard(
    api_key="your-api-key",
    ssa_url="https://ssa.agentguard.io"
)

# Secure your agent tool calls
result = await guard.execute_tool(
    tool_name="web-search",
    parameters={"query": "AI agent security"},
    context={"session_id": "user-session-123"}
)

print(f"Secure result: {result.data}")
print(f"Security decision: {result.security_decision}")

✨ Features

• 🛡️ Runtime Security Enforcement - Mediate all agent tool/API calls through security policies
• 📋 Policy-Based Access Control - Define and enforce security policies with ease
• 🔍 Comprehensive Audit Trails - Track every agent action with tamper-evident logs
• ⚡ High Performance - <100ms latency for security decisions
• 🔧 Type Hints - Full type annotations for better IDE support
• 🎯 Request Transformation - Automatically transform risky requests into safer alternatives
• 🔐 Zero-Trust Architecture - Never trust, always verify
• 📊 Real-time Monitoring - Track agent behavior and security events
• 🔄 Async Support - Built-in async/await support for modern Python applications

📖 Installation

Using pip

pip install agentguard-sdk

Using poetry

poetry add agentguard-sdk

From source

git clone https://github.com/agentguard-ai/agentguard-python.git
cd agentguard-python
pip install -e .

🎯 Usage Examples

Basic Usage

from agentguard import AgentGuard

guard = AgentGuard(
    api_key="your-api-key",
    ssa_url="http://localhost:3000"
)

# Synchronous execution
result = guard.execute_tool_sync(
    tool_name="file-write",
    parameters={
        "path": "/data/output.txt",
        "content": "Agent generated content"
    },
    context={
        "session_id": "agent-session-456",
        "user_id": "user-123"
    }
)

if result.success:
    print(f"Tool executed securely: {result.data}")
else:
    print(f"Security policy blocked: {result.error}")

Async Usage

import asyncio
from agentguard import AgentGuard

async def main():
    guard = AgentGuard(
        api_key="your-api-key",
        ssa_url="http://localhost:3000"
    )
    
    result = await guard.execute_tool(
        tool_name="database-query",
        parameters={"query": "SELECT * FROM users LIMIT 10"},
        context={"session_id": "session-789"}
    )
    
    print(result.data)

asyncio.run(main())

Policy Testing

from agentguard import PolicyTester

tester = PolicyTester()

# Test your policies before deployment
result = tester.test_policy(
    policy=my_policy,
    request={
        "tool_name": "database-query",
        "parameters": {"query": "SELECT * FROM users"}
    }
)

print(f"Policy decision: {result.decision}")
print(f"Reasoning: {result.reason}")

Policy Builder

from agentguard import PolicyBuilder

policy = (
    PolicyBuilder()
    .name("restrict-file-operations")
    .description("Prevent file write operations")
    .add_rule(
        condition={"tool_name": "file-write"},
        action="deny",
        reason="File write operations are not allowed"
    )
    .add_rule(
        condition={"tool_name": "file-read"},
        action="allow",
        reason="File read operations are permitted"
    )
    .build()
)

print(f"Policy created: {policy}")

🔧 Configuration

Basic Configuration

guard = AgentGuard(
    api_key="your-api-key",
    ssa_url="https://ssa.agentguard.io",
    timeout=5.0,
    max_retries=3
)

Advanced Configuration

guard = AgentGuard(
    api_key=os.getenv("AGENTGUARD_API_KEY"),
    ssa_url=os.getenv("AGENTGUARD_SSA_URL"),
    
    # Timeout settings
    timeout=10.0,
    max_retries=3,
    retry_delay=1.0,
    
    # Logging
    log_level="INFO",
    
    # Custom headers
    headers={
        "X-Custom-Header": "value"
    },
    
    # Callback hooks
    on_security_decision=lambda decision: print(f"Decision: {decision}"),
    on_error=lambda error: print(f"Error: {error}")
)

📚 Documentation

• Getting Started Guide
• API Reference
• Policy Configuration
• Examples

🤝 Contributing

We welcome contributions! Please see our Contributing Guide for details.

📄 License

MIT © AgentGuard

🔒 Security

Security is our top priority. If you discover a security vulnerability, please email [e-mail] instead of using the issue tracker.

See SECURITY.md for more details.

🌟 Why AgentGuard?

The Problem

AI agents are powerful but pose significant security risks:

• Unrestricted access to tools and APIs
• No audit trail of agent actions
• Difficult to enforce security policies
• Hard to debug agent behavior

The Solution

AgentGuard provides:

• ✅ Runtime Security - Every tool call is evaluated before execution
• ✅ Policy Enforcement - Define what agents can and cannot do
• ✅ Audit Trails - Complete visibility into agent actions
• ✅ Request Transformation - Automatically make risky requests safer
• ✅ Zero-Trust - Never trust, always verify

🚀 Roadmap

• Core SDK with policy enforcement
• Type hints and async support
• Comprehensive test suite
• Drop-in integrations (LangChain, CrewAI, AutoGPT)
• Built-in guardrails library
• Cost monitoring and budget enforcement
• Visual policy management UI
• Real-time monitoring dashboard

💬 Community

• GitHub Discussions - Ask questions and share ideas
• GitHub Issues - Report bugs and request features
• Email - Direct contact

---

Built with ❤️ by the AgentGuard team

GitHub • PyPI • Email