AgentShield: Python Firewall for AI Agent Spend Control

Introduction

AgentShield is a pure Python library designed to act as a firewall for AI agent spending. It helps stop runaway AI agents before they consume excessive budget by offering 10 composable spend rules, evaluated per-transaction in under 1ms. While the project's primary development has been consolidated into sipi.bot, the AgentShield Python package and its test fixtures remain available for existing users and open-source contributions.

Why Use & Benefits

AI agents, while powerful, can sometimes incur unexpected costs. AgentShield provides a robust solution to manage and control these expenditures with its lightweight, high-performance engine. Key benefits include its pure Python 3.11 stdlib implementation with zero dependencies, ensuring easy integration and minimal overhead. It uses decimal.Decimal for precise money calculations, operates statelessly, and guarantees deterministic results, all while evaluating transactions in less than 1ms. The library offers a comprehensive set of 10 rule types, from transaction_limit and daily_total to velocity and merchant_allowlist, allowing fine-grained control over agent spending.

Installation

To get started with AgentShield, install the package via pip:

pip install agentshield-spend

Note: The import name for the library is agentshield, as the PyPI name agentshield belongs to an unrelated project.

Examples

Here's a quick example demonstrating how to use AgentShield to evaluate a transaction against defined spend rules:

from agentshield import SpendControlEngine

engine = SpendControlEngine()

# A transaction your agent wants to make
transaction = {
    "amount": 500.00,
    "merchant": "openai-api",
    "category": "llm_inference",
    "agent_id": "my-agent",
    "timestamp": "2026-08-10T10:00:00Z",
}

# Your spend-control rules
rules = [
    {"id": "r1", "type": "transaction_limit", "priority": 1,
     "params": {"max_amount": 250}, "action": "BLOCK"},
    {"id": "r2", "type": "daily_total", "priority": 2,
     "params": {"max_daily": 2000}, "action": "BLOCK"},
    {"id": "r3", "type": "velocity", "priority": 3,
     "params": {"window_minutes": 60, "max_count": 10}, "action": "FLAGGED"},
]

# Prior transactions today (for daily_total and velocity checks)
prior_transactions = []

# Evaluate, returns in <1ms
result = engine.evaluate(transaction, rules, prior_transactions)
print(result["decision"])  # BLOCKED
print(result["reason"])    # Transaction amount $500.00 exceeds limit of $250.00

Links

For more information and to explore the project further, refer to these links:

• Canonical product
• Paid implementation pilot
• sipi.bot public eval report
• GitHub Repository