Humanoid Agents is a platform for simulating generative agents that exhibit human-like thinking, talking, and behavior. It targets researchers and developers interested in creating more realistic AI agents by incorporating System 1 thinking, including basic needs, emotions, and relationships, alongside System 2 logical reasoning. The platform offers a simulation engine, an analytics dashboard, and a Unity WebGL game interface for visualization.

How It Works

The platform simulates agents by initializing them with seed information, planning their daily activities, and executing actions. Agents can converse with each other, influencing their relationships. Actions are evaluated for their impact on the agent's basic needs and emotional state, which in turn informs future planning. This approach aims to imbue agents with a more nuanced, human-like decision-making process driven by internal states.

Quick Start & Requirements

• Install via git clone https://github.com/HumanoidAgents/HumanoidAgents.git and pip install -e .
• Requires Python and an OpenAI API key for LLM access (or configuration for local LLMs via an OpenAI-compatible server like LM Studio).
• Estimated cost for 2-3 agents per simulated day is $2-5, taking 45-60 minutes.
• Official Demo: humanoidagents.com
• YouTube Walkthrough: [Link not provided in README, but mentioned]

Highlighted Details

• Supports simulation of specific fictional characters (e.g., Big Bang Theory, Friends).
• Offers a Flask-based REST API for real-time interaction and server-client mode.
• Allows customization of agent basic needs and extension of the HumanoidAgent class for deeper behavioral modeling.
• Includes an analytics dashboard for visualizing agent status, needs, and relationships over time.

Maintenance & Community

• Accepted to EMNLP System Demonstrations 2023.
• Project website: humanoidagents.com
• Code repository: github.com/HumanoidAgents/HumanoidAgents

Licensing & Compatibility

• The README does not explicitly state a license. The code repository should be checked for license files.

Limitations & Caveats

• Customization of locations and agents for the Unity WebGL Game interface is noted as "coming soon."
• Using OpenAI incurs costs, and local LLM setup requires an OpenAI-compatible server.
• The global start date for user-defined memories is fixed at 2023-01-02, requiring code adjustment for earlier dates.