Agentarium

Introduction to Agentarium

Agentarium is an open-source Python framework designed to facilitate the creation and management of simulations populated with autonomous, AI-powered agents. This flexible and intuitive platform enables users to design complex, interactive environments where agents can act, learn, and evolve through dynamic interactions and evolving behavior.

Features

• Advanced Agent Management: Agentarium allows the generation and orchestration of multiple AI agents, each with distinct roles and capabilities, making it ideal for simulations with diverse agent characters and complex ecosystems.

• Robust Interaction Management: The framework includes sophisticated tools for coordinating complex interactions among agents, enabling nuanced communication patterns and multi-agent cooperation or competition scenarios.

• Checkpoint System: Agents' states and interactions can be saved and restored, facilitating iterative testing, debugging, and scenario development.

• Data Generation: Users can generate synthetic data through agent interactions, providing invaluable datasets for various analytical purposes such as machine learning experiments or statistical analysis.

• Performance Optimization: Built with efficiency and scalability in mind, Agentarium supports expansive simulations with minimal performance overhead.

• Flexible Environment Configuration: Easily configure custom environments using YAML files, allowing tailored agent experiences and experiments.

• Extensible Architecture: The design promotes easy extension and customization, making the framework adaptable to specific needs or research areas.

Usage Examples

Basic Chat Interaction

In this simple example, create a basic chat interaction between two agents:

from agentarium import Agent

alice = Agent.create_agent()
bob = Agent.create_agent()

alice.talk_to(bob, "Hello Bob! I heard you're working on some interesting data science projects.")
bob.act()

Synthetic Data Generation

Demonstrate how agents can generate synthetic data:

from agentarium import Agent
from agentarium.CheckpointManager import CheckpointManager

checkpoint = CheckpointManager("demo")

alice = Agent.create_agent()
bob = Agent.create_agent()

alice.talk_to(bob, "What a beautiful day!")
checkpoint.update(step="interaction_1")

checkpoint.save()

More examples are available in the examples/ directory.

Documentation & Configuration

Agentarium supports configuration through YAML files, making it straightforward to set up agent environments tailored to specific needs:

llm:
  provider: "openai" # any provider supported by aisuite
  model: "gpt-4o-mini" # any model supported by the provider

aisuite: # optional, credentials for aisuite
  openai:
    api_key: "sk-..."

Key Components

• Agent: The fundamental class for creating AI-driven agents.
• CheckpointManager: Facilitates the saving and loading of agent states.

Agentarium serves as an invaluable tool for researchers, educators, and developers involved in AI, machine learning, or any field requiring sophisticated simulation environments with autonomous agents.