SteerPlane provides a robust runtime control plane for autonomous AI agents, designed to enhance operational safety and efficiency through cost limits, loop detection, and comprehensive observability—all integrated with a simple decorator.

Key Features

SteerPlane addresses critical challenges faced by AI agents that interact with APIs, execute code, and make complex decisions. Without proper guardrails, errors can lead to infinite loops, excessive API costs, and unmonitored actions. SteerPlane mitigates these risks with:

• Loop Detection: Employs a sliding-window pattern to monitor and prevent repetitive agent behavior in real time.
• Hard Cost Ceiling: Enables configuration of a maximum allowable cost per run. If the total cost exceeds the specified limit, execution is halted immediately.
• Step Limit Enforcement: Caps the maximum execution steps, safeguarding against unbounded resource consumption.
• Deep Telemetry: Automatically captures detailed metrics for each execution step, including action name, token usage, cost, latency, and overall status.
• Real-Time Dashboard: An interactive Next.js dashboard with auto-refresh capabilities, allowing users to observe agent performance and cost breakdowns in real time.
• Graceful Degradation: Ensures agents remain protected, utilizing local enforcement of limits if external APIs become unavailable.

How to Use

Integrating SteerPlane requires just one line of code. Here is an example to illustrate its implementation:

from steerplane import guard

@guard(
    agent_name="support_bot",
    max_cost_usd=10.00,
    max_steps=50,
    detect_loops=True
)
def run_agent():
    # Agent logic here
    agent.run()

With the above decorator, agents are monitored continuously through their execution cycle. Should an agent exceed its configured limits, it will be terminated automatically, offering peace of mind while optimizing resource management.

Supported Technologies

SteerPlane is designed for use with:

• Python 3.10+
• Node.js 18+

Documentation and Community

Extensive documentation is available to assist users at every step, from installation to advanced usage. Contributions are encouraged, and a welcoming community is ready to support developers through the integration of SteerPlane into their own projects.

For more details, see the documentation and explore the features that make SteerPlane a crucial tool for managing autonomous AI agents effectively.