AI Town

What is AI Town?

AI Town is an open-source virtual simulation thing where AI characters hang around, talk, mingle, and then somehow develop evolving relationships over time. It was made with help from Convex and Andreessen Horowitz, and you can kind of trace the inspiration back to Stanford’s Generative Agents research. In practice it feels like a deployable starter kit, so developers can tailor it into their own AI-powered virtual world, multiplayer setup, or even a full-on agent simulation. AI Town uses Convex as its game engine, along with the database, vector search, and a shared global state that ties everything together. Then developers can plug in models like Ollama, OpenAI, Together AI, or other providers that are OpenAI-compatible, and they can customize characters, maps, storylines, and the entire gameplay mechanics loop.

AI Town launched in August 2023 and quickly became a notable open-source AI project. Its GitHub repository currently shows about 9,900 stars, more than 1,100 forks, and over 80 watchers. The project is MIT licensed and was designed as a JavaScript- and TypeScript-based alternative to Python-focused agent simulations. It supports multiplayer-style interactions, persistent memories through vector search, customizable characters, maps, stories, and multiple LLM providers. The project can run locally, through Docker, or on Convex cloud infrastructure.

• Founder: Yoko Li was closely involved in the original AI Town concept, which originated from her companion application project. The broader project was developed through collaboration involving Andreessen Horowitz and Convex, with James Cowling and other contributors playing important technical roles.
• Launch: August 2023. AI Town was publicly introduced as an open-source project built through collaboration between A16Z and Convex.
• Use Cases
  • Building autonomous AI character simulations
  • Creating multiplayer AI games
  • Experimenting with multi-agent systems
  • Researching AI social interactions
  • Developing virtual worlds populated by AI characters
  • Testing long-term AI memory and conversations
  • Creating custom AI communities and interactive experiences
  • Learning how to build agent-based applications with JavaScript and TypeScript
• Technology
  • Convex for the backend, database, game engine, transactions, and vector search
  • JavaScript and TypeScript for application development
  • Ollama for local LLM inference
  • Llama 3 as the default chat model in the project
  • mxbai-embed-large for default embeddings
  • OpenAI and Together AI support
  • OpenAI-compatible APIs, including providers such as Anthropic, Groq, and Azure
  • PixiJS for rendering and interactive experiences
  • Replicate and MusicGen for optional background music generation
  • Clerk for optional authentication
• Target Users
  • AI developers
  • Game developers
  • Researchers
  • AI startups
  • JavaScript and TypeScript developers
  • Students experimenting with AI agents
  • Developers building multiplayer applications
  • Teams exploring generative AI simulations
• Acquisition: AI Town has not been acquired. It remains an open-source project associated with Convex and the a16z infrastructure ecosystem.

AI Town Key Features

AI Town's key features are

• AI Character Simulation: AI Town creates a virtual environment where autonomous AI characters can move around, communicate, socialize, and develop relationships.
• Persistent AI Memories: Characters can use stored memories to influence later conversations, making interactions more continuous rather than treating every conversation as completely new.
• Custom Characters and Stories: Developers can change character identities, descriptions, stories, sprites, and behaviors to create their own simulated world.
• Multiplayer Interaction: Users can walk around the virtual environment and interact with AI citizens, including inviting nearby characters into conversations.
• Shared Global State: Convex provides shared state, transactions, and a simulation engine designed to support interactive multiplayer experiences.
• Vector Search: AI Town uses vector search to retrieve relevant memories and support more context-aware character conversations.
• Multiple LLM Providers: Developers are not locked into a single model provider. The project supports Ollama, OpenAI, Together AI, and other OpenAI-compatible APIs.
• Custom Maps: Developers can modify the environment and create different virtual locations using map and tileset assets.
• Open Source Architecture: The MIT-licensed codebase can be inspected, modified, deployed, and extended for different AI experiments and applications.

AI Town Pricing

AI Town offers various pricing plans

• AI Town: Free and open source under the MIT license.
• Convex Free Plan: $0 per month, suitable for development and experimentation.
• Convex Professional: $25 per developer per month.
• Business/Enterprise: Custom pricing, with larger production requirements and support.
• LLM costs: Using paid providers such as OpenAI or Together AI may add separate API charges.
• Local models: Developers can use Ollama to run models locally, avoiding API charges but requiring suitable hardware.
• Hosting: Additional infrastructure costs may apply depending on where and how AI Town is deployed.

Disclaimer: For the latest and most accurate pricing information, please visit the official AI Town website.

Who is Using AI Town?

A diverse range of users and organizations utilize AI Town

• AI developers experimenting with autonomous agents
• Open-source developers building AI simulations
• Game developers exploring AI-driven gameplay
• Researchers studying multi-agent behavior
• Developers creating AI companions and virtual communities
• Students learning about agent architectures
• Startups experimenting with AI-native interactive products
• Open-source contributors extending the project

What Are the Best AI Town Alternatives?

Some AI Town alternatives are

• Convai
• Inworld
• Simulacra
• AgentVerse
• Generative Agents
• AutoGen
• CrewAI
• LangGraph

Pros and Cons

Pros

• Open source and MIT licensed
• Highly customizable
• Supports autonomous AI characters
• Designed for multi-agent experimentation
• Supports persistent memories
• Uses real-time backend infrastructure
• Supports multiple LLM providers
• Can run locally or in the cloud
• Suitable for developers who want to modify the underlying architecture
• Strong foundation for building custom AI simulations

Cons

• Requires development knowledge
• Setup can be technically demanding for beginners
• LLM usage can create additional costs
• Character behavior can be unpredictable
• Local AI inference requires suitable hardware
• It is more of a development framework than a finished consumer product
• Customization requires working with the codebase and project data

Why Choose AI Town?

• Open Source Development: AI Town provides developers with an MIT-licensed codebase that they can modify and deploy for their own projects.
• Strong AI Experimentation Platform: It provides a practical environment for testing autonomous agents, memories, social interactions, and multi-agent behavior.
• Flexible Model Support: Developers can choose between local inference and multiple cloud-based LLM providers.
• Customizable Worlds: Characters, stories, maps, sprites, and gameplay mechanics can be modified to create unique simulations.
• Real-Time Architecture: Convex provides the shared state, database, transactions, and simulation infrastructure required for interactive applications.
• Educational Value: Developers can study a real JavaScript and TypeScript implementation of an AI agent simulation rather than relying only on theoretical research.

AI Town vs Competitors

The main difference between AI Town, Generative Agents, Inworld AI, and Convai is that AI Town combines an open-source virtual world, autonomous characters, persistent memories, multiplayer interaction, and a customizable backend into one developer-focused starter kit. Generative Agents is primarily research, while Inworld AI and Convai focus more on production-ready character experiences. AI Town is therefore particularly attractive to developers who want to inspect, modify, and extend the underlying system.

How Do We Rate AI Town?

AI Town Review

AI Town is one of the more interesting open-source projects for developers who want to understand how autonomous AI characters can interact inside a shared environment. Its biggest strength is flexibility. Instead of presenting a fixed AI experience, it gives developers a foundation they can reshape. The combination of Convex, vector search, persistent memories, configurable LLM providers, and customizable characters makes the project useful for experimentation as well as learning. Its open-source nature also makes it easier to inspect how the system works. However, AI Town is not designed as a no-code AI tool. Developers need to be comfortable with JavaScript or TypeScript, backend configuration, LLM providers, and deployment. For technical users, that complexity is part of its appeal. For non-technical users, ready-made AI character platforms may be easier. Overall, AI Town earns a 4.5 out of 5 because it offers a strong combination of openness, flexibility, AI experimentation, and extensibility.

Conclusion

AI Town is much more than just an AI game or a virtual world, really. It is an open-source foundation for trying out autonomous characters, lasting memories, social behavior, and multi-agent systems in a more hands-on way. The convex-powered architecture, its flexible support for large language models, and the ability to craft customizable characters, plus extensible gameplay parts, make it particularly helpful for developers as well as researchers. Even if the technical setup can feel a bit rough for beginners, more seasoned developers can treat AI Town as a base layer for games, simulations, experiments, and other AI-native experiences. And since it’s open source, it has an extra edge compared to those closed AI character platforms.