Starter Kit

GitHub

Supervisors are assigned to monitor tools that the agent can use, and block bad executions

Asteroid helps you supervise and monitor your AI agents, enabling safe and reliable deployment in any domain

Hook your agent up to Asteroid in minutes

Walk through setting up an agent that uses Asteroid, from scratch

Wrap your agent tools in our decorator to enable supervision

A human supervisor lets you easily add Human-in-the-Loop to your AI agent

LLM Supervisors allow you to use LLM-as-a-Judge to monitor your agent behaviour

Custom supervisors let you run arbitrary code to check on agent actions before they execute

Integrate Asteroid supervision into OpenAI-powered applications

Supervise and evaluate agentic systems using UK's AI Safety Institute's Inspect AI evaluation framework