Welcome to Meta Agents Research Environments¶

Meta Agents Research Environments (ARE) is a dynamic simulation research platform for training and evaluating AI agents on complex, multi-step tasks that mirror real-world challenges. Unlike static benchmarks, ARE creates evolving environments where agents must adapt their strategies as new information emerges and conditions change over time. In particular, ARE runs the Gaia2 Benchmark, a follow-up to Gaia, evaluating a broader range of agent capabilities.

What Can ARE Do?¶

Dynamic Simulations: Create realistic scenarios that evolve over minutes, hours, or days - simulating complex workflows that require persistent reasoning and adaptation.
Agent Evaluation: Test AI agents on multi-step tasks with comprehensive benchmarking tools, including the Gaia2 benchmark with 800 scenarios across 10 universes.
Interactive Applications: Agents interact with realistic apps like email, calendars, file systems, and messaging - each with domain-specific data and behaviors.
Research & Benchmarking: Systematic evaluation with parallel execution, multiple model support, and automatic result collection for the research community.

Next Steps¶

Quick Start

For a step-by-step guide to use ARE to evaluation your agents, see the Quick Start page.

Foundations

Learn More: Dive deeper into the core concepts of agents, environments, apps, events, and scenarios.

Demo

Try the ARE Demo on Hugging Face — Play around with the agent platform directly in your browser, no installation required!

Gaia2 and Leaderboard Submission

Build and evaluate your agents on the Gaia2 benchmark, a comprehensive suite of 800 dynamic scenarios across 10 universes.

Working with Scenarios

Explore other ways to create and work with scenarios, including using the CLI, Python, and JSON.

Explore Code

Explore the ARE code on GitHub.

Documentation Structure¶

User Guide

API Reference

Indices and tables¶

Index