Comparison of the four major Crypto X AI frameworks: ELIZA, GAME, ARC and ZEREPY

Author: Deep Value Memetics, Translation: Jinse Finance xiaozou

In this article, we will explore the prospects of Crypto X AI frameworks. We will focus on the four major frameworks (ELIZA, GAME, ARC, ZEREPY) and their respective technical differences.

1. Introduction

We have studied and tested the four major Crypto X AI frameworks, ELIZA, GAME, ARC, and ZEREPY, in the past week, and our conclusions are as follows.

We believe that AI16Z will continue to dominate. Eliza’s value (~60% market share, over $1 billion market cap) lies in its first-mover advantage (Lindy effect) and its growing adoption by developers, as evidenced by 193 contributors, 1,800 forks, and over 6,000 stars, making it one of the most popular repositories on Github.

GAME (~20% market share, ~$300 million market cap) has been very successful so far and is gaining rapid adoption, as VIRTUAL just announced that the platform has over 200 projects, 150,000 daily requests, and 200% week-over-week growth. GAME will continue to benefit from VIRTUAL’s rise and will be one of the biggest winners in its ecosystem.

Rig (ARC, ~15% market share, ~$160 million market cap) is very compelling because its modular design is very easy to operate and can dominate the Solana ecosystem (RUST) as a “pure-play”.

Zerepy (~5% market share, ~$300M market cap) is a relatively niche application that caters to the enthusiastic ZEREBRO community, and its recent collaboration with the ai16z community may have synergies.

We note that our market share calculations cover market capitalization, development record, and underlying operating system end markets.

We believe that the framework segment will be the fastest growing area in this market cycle, and the total market capitalization of $1.7 billion may easily grow to $20 billion, which is still relatively conservative compared to the peak valuations of L1s in 2021, when many L1s reached valuations of $20 billion+. While these frameworks all serve different end markets (chain/ecosystem), given that we believe the space is in an upward trend, a market cap weighted approach may be the most prudent approach.

2. Four major frameworks

In the following table, we list the key technologies, components and advantages of each major framework.

(1) Framework Overview

In the intersection of AI and Crypto, there are several frameworks that promote the development of AI. They are ELIZA from AI16Z, RIG from ARC, ZEREBRO from ZEREPY, and VIRTUAL from GAME. Each framework addresses different needs and philosophies in the development of AI agents, ranging from open source community projects to enterprise-level solutions that focus on performance.

This article first introduces the frameworks, telling you what they are, what programming languages ​​they use, technical architectures, algorithms, what unique features they have, and what potential use cases the frameworks can be used for. Then, we compare each framework in terms of usability, scalability, adaptability, and performance, exploring the strengths and limitations of each.

ELIZA (developed by ai16z)

Eliza is an open source framework for multi-agent simulation designed to create, deploy, and manage autonomous AI agents. Developed in the TypeScript programming language, it provides a flexible and extensible platform for building intelligent agents that are able to interact with humans on multiple platforms and maintain consistent personality and knowledge.

The core features of the framework include a multi-agent architecture that supports the simultaneous deployment and management of multiple unique AI personalities, a role system for creating different agents using a role file framework, and memory management capabilities that provide long-term memory and context awareness through an advanced retrieval augmentation generation (RAG) system. In addition, the Eliza framework provides smooth platform integration and reliable connection with Discord, X, and other social media platforms.

From the communication and media capabilities of AI agents, Eliza is an excellent choice. In terms of communication, the framework supports integration with Discord's voice channel function, X functions, Telegram, and direct API access for customized use cases. On the other hand, the media processing capabilities of the framework can be extended to PDF document reading and analysis, link content extraction and summarization, audio transcription, video content processing, image analysis, and conversation summarization, which can effectively handle various types of media input and output.

The Eliza framework provides flexible AI model support through local inference of open source models, cloud inference from OpenAI, and default configurations such as Nous Hermes Llama 3.1B, and integrates support for Claude to handle complex tasks. Eliza adopts a modular architecture with a wide range of operating systems, custom client support, and comprehensive APIs, ensuring scalability and adaptability between applications.

Eliza's use cases span multiple areas, such as: AI assistants for customer support, community moderation, and personal tasks, as well as social media roles such as content automatic creators, interactive robots, and brand representatives. It can also act as a knowledge worker, playing roles such as research assistants, content analysts, and document processors, and supports interactive roles in the form of role-playing robots, educational mentors, and entertainment agents.

Eliza's architecture is built around an agent runtime that seamlessly integrates with its role system (supported by model providers), memory manager (connected to the database), and operating system (linked to the platform client). The framework’s unique features include a plugin system that enables modular functionality expansion, support for multimodal interactions such as voice, text, and media, and compatibility with leading AI models such as Llama, GPT-4, and Claude. With its versatile functionality and powerful design, Eliza stands out as a powerful tool for developing AI applications across domains.

G.A.M.E (Developed by Virtuals Protocol)

The Generative Autonomous Multimodal Entity Framework (G.A.M.E) aims to provide developers with API and SDK access to conduct AI agent experiments. This framework provides a structured approach to manage the behavior, decision-making, and learning process of AI agents.

The core components are as follows:First, the Agent Prompting Interface is the entry point for developers to integrate GAME into their agents to access agent behaviors. The Perception Subsystem starts a session by specifying parameters such as the session ID, agent ID, user, and other relevant details.

It synthesizes incoming information into a format suitable for the Strategic Planning Engine (Strategic Planning Engine) to act as a sensory input mechanism for AI agents, whether in the form of dialogue or reaction. At its core is the dialogue processing module, which is used to process messages and responses from agents and collaborate with the perception subsystem to effectively interpret and respond to input.

The strategic planning engine works with the dialogue processing module and the on-chain wallet operator to generate responses and plans. The engine functions at two levels: as a high-level planner, it creates a wide range of strategies based on context or goals; as a low-level strategy, it converts these strategies into actionable strategies, which are further divided into action planners for specifying tasks and plan executors for executing tasks.

Another independent but important component is the World Context, which references the environment, global information, and game state to provide the necessary context for the agent's decision-making. In addition, the Agent Repository is used to store long-term attributes such as goals, reflections, experience, and personality, which together shape the agent's behavior and decision-making process.

The framework uses short-term working memory and long-term memory processors. Short-term memory retains information about previous actions, outcomes, and current plans. In contrast, long-term memory processors extract key information based on criteria such as importance, recency, and relevance. Long-term memory stores knowledge such as the agent's experience, reflection, dynamic personality, world context, and working memory to enhance decision-making and provide a foundation for learning.

The learning module uses data from the perception subsystem to generate general knowledge, which is fed back into the system to improve future interactions. Developers can input feedback about actions, game states, and sensory data through the interface to enhance the AI ​​agent's learning ability and improve its planning and decision-making capabilities.

The workflow begins with developers interacting through the agent prompt interface. The input is processed by the perception subsystem and forwarded to the dialogue processing module, which manages the interaction logic. The strategic planning engine then develops and executes plans based on this information, utilizing high-level strategies and detailed action plans.

Data from the world context and agent repository inform these processes, while working memory keeps track of immediate tasks. Meanwhile, the long-term memory processor stores and retrieves long-term knowledge. The learning module analyzes the results and integrates new knowledge into the system, enabling continuous improvement of the agent's behavior and interactions.

RIG (Developed by ARC)

Rig is an open source Rust framework designed to simplify the development of large language model applications. It provides a unified interface for interacting with multiple LLM providers such as OpenAI and Anthropic, and supports a variety of vector stores, including MongoDB and Neo4j. The framework's modular architecture is unique in its core components such as the Provider Abstraction Layer, vector store integration, and agent system to facilitate seamless interaction with LLMs.

Rig's primary audience includes developers building AI/ML applications with Rust, and secondarily, organizations seeking to integrate multiple LLM providers and vector stores into their own Rust applications. The repository uses a workspace architecture with multiple crates, supporting extensibility and efficient project management. Its key features include a provider abstraction layer, which provides standardization for completing and embedding APIs between different LLM providers with consistent error handling. The Vector Store Integration component provides an abstract interface for multiple backends and supports vector similarity search. The proxy system simplifies LLM interactions, supports retrieval enhancement generation (RAG) and tool integration. In addition, the embedding framework provides batch processing capabilities and type safety embedding operations.

Rig takes advantage of multiple technical advantages to ensure reliability and performance. Asynchronous operations leverage Rust's asynchronous runtime to efficiently handle a large number of concurrent requests. The framework's inherent error handling mechanism improves resilience to failures in AI provider or database operations. Type safety prevents errors during compilation, thereby enhancing the maintainability of the code. Efficient serialization and deserialization processes support data processing in formats such as JSON, which is critical for AI service communication and storage. Detailed logging and instrumentation further help debug and monitor applications.

Rig’s workflow begins when a client initiates a request that interacts with the appropriate LLM model through a provider abstraction layer. Data is then processed by the core layer, where agents can use tools or access a vector store of context. Responses are generated and refined through complex workflows (such as RAG) involving document retrieval and context understanding before being returned to the client. The system integrates with multiple LLM providers and vector stores and is adaptive to model availability or performance updates.

Rig’s use cases are diverse and include question-answering systems that retrieve relevant documents to provide accurate responses, document search and retrieval systems for efficient content discovery, and chatbots or virtual assistants that provide context-aware interactions for customer service or education. It also supports content generation, enabling the creation of text and other materials based on learning patterns, making it a versatile tool for developers and organizations.

Zerepy (by ZEREPY and blorm)

ZerePy is an open source framework written in Python designed to deploy agents on X using OpenAI or Anthropic LLM. Derived from a modular version of the Zerebro backend, ZerePy allows developers to launch agents with similar functionality to the core Zerebro. While the framework provides a foundation for agent deployment, fine-tuning the model is essential to generate creative output. ZerePy simplifies the development and deployment of personalized AI agents, especially for content creation on social platforms, fostering an AI-driven creative ecosystem for art and decentralized applications.

The framework is written in Python, emphasizes agent autonomy, and focuses on creative output generation, in line with the architecture of and partnership with ELIZA. Its modular design supports in-memory system integration and supports the deployment of agents on social platforms. Key features include a command-line interface for agent management, integration with Twitter, support for OpenAI and Anthropic LLM, and a modular connection system for enhanced functionality.

ZerePy's use cases cover the field of social media automation, where users can deploy AI agents to post, reply, like, and retweet, thereby increasing platform engagement. In addition, it caters to content creation in areas such as music, memes, and NFTs, making it an important tool for digital art and blockchain-based content platforms.

(2) Comparison of the four major frameworks

In our view, each framework provides a unique approach to AI development that meets specific needs and environments, and we shift the focus from the competitive relationship of these frameworks to the uniqueness of each framework.

ELIZA stands out for its user-friendly interface, especially for developers familiar with JavaScript and Node.js environments. Its comprehensive documentation helps set up AI agents on a variety of platforms, although its extensive feature set may bring a certain learning curve. Developed in TypeScript, Eliza is ideal for building agents embedded in the web, as most web infrastructure frontends are developed in TypeScript. The framework is known for its multi-agent architecture, which allows different AI personalities to be deployed on platforms such as Discord, X, and Telegram. Its advanced memory management RAG system makes it particularly effective for AI assistants in customer support or social media applications. While it offers flexibility, strong community support, and consistent cross-platform performance, it is still in its early stages and may pose a learning curve for developers.

GAME is designed for game developers, providing a low-code or no-code interface through an API, making it accessible to less technical users in the gaming field. However, its focus on game development and blockchain integration may pose a steep learning curve for those without relevant experience. It excels in procedural content generation and NPC behavior, but is limited by the complexity added by its niche and blockchain integration.

Due to its use of the Rust language, Rig may not be very user-friendly given the complexity of the language, which poses a significant learning challenge, but it has intuitive interactions for those who are proficient in system programming. The programming language itself is known for its performance and memory safety compared to typescripe. It has strict compile-time checking and zero-cost abstractions, which are necessary to run complex AI algorithms. The language is very efficient, and its low degree of control makes it ideal for resource-intensive AI applications. The framework provides a high-performance solution with a modular and extensible design, making it ideal for enterprise applications. However, for developers who are not familiar with Rust, using Rust cannot avoid facing a steep learning curve.

ZerePy leverages Python, providing high usability for creative AI tasks, with a low learning curve for Python developers, especially for those with an AI/ML background, and benefits from strong community support due to Zerebro's crypto community. ZerePy excels in creative AI applications such as NFTs, positioning itself as a powerful tool for digital media and art. While it thrives on creativity, its scope is relatively narrow compared to other frameworks.

In terms of scalability, ELIZA has made great strides in its V2 update, which introduced a unified message line and an extensible core framework that supports effective management across multiple platforms. However, if not optimized, the management of such multi-platform interactions may pose scalability challenges.

GAME excels in the real-time processing required for games, and scalability is managed through efficient algorithms and underlying blockchain distributed systems, although it may be limited by specific game engines or blockchain networks.

The Rig framework leverages Rust's scalability performance and is designed for high-throughput applications, which is particularly effective for enterprise-level deployments, although this may mean that achieving true scalability requires complex settings.

Zerepy's scalability is oriented towards creative output and is supported by community contributions, but its focus may limit its application in a wider range of AI environments, and scalability may be tested by the diversity of creative tasks rather than the number of users.

In terms of adaptability, ELIZA leads with its plugin system and cross-platform compatibility, and its GAME for gaming environments and Rig for handling complex AI tasks also excel. ZerePy shows high adaptability in the creative field, but is less suitable for broader AI applications.

In terms of performance, ELIZA is optimized for fast social media interactions, where fast response times are key, but its performance may vary when handling more complex computing tasks.

GAME, developed by Virtual Protocol, focuses on high-performance real-time interactions in gaming scenarios, leveraging efficient decision-making processes and potential blockchains for decentralized AI operations.

The Rig framework is based on the Rust language and provides excellent performance for high-performance computing tasks, suitable for enterprise applications where computing efficiency is critical.

Zerepy's performance is tailored for the creation of creative content, and its metrics are centered on the efficiency and quality of content generation, which may not be very universal outside the creative field.

ELIZA's strengths are flexibility and extensibility, making it highly adaptable through its plugin system and role configuration, facilitating cross-platform social AI interactions.

GAME provides unique real-time interaction capabilities in games, enhancing novel AI engagement through blockchain integration.

Rig's strengths are its performance and extensibility for enterprise AI tasks, with a focus on clean, modular code for long-term project health.

Zerepy excels at fostering creativity, leading the way in AI applications for digital art, and supported by a vibrant community-driven development model.

Each framework has its own limitations. ELIZA is still in its early stages, with potential stability issues and a learning curve for new developers. Niche games may limit wider applications, and blockchain also adds complexity. Rig's steep learning curve due to Rust may scare off some developers, and Zerepy's narrow focus on creative output may limit its use in other AI fields.

(3) Framework comparison summary

Rig (ARC):

Language: Rust, focusing on security and performance.

Use case: Ideal for enterprise-level AI applications because it focuses on efficiency and scalability.

Community: Less community-driven, more focused on technical developers.

Eliza (AI16Z):

Language: TypeScript, emphasizing web3 flexibility and community engagement.

Use Cases: Designed for social interactions, DAOs, and transactions, with a particular emphasis on multi-agent systems.

Community: Highly community-driven, with extensive GitHub participation.

ZerePy (ZEREBRO):

Language: Python, making it accessible to a wider base of AI developers.

Use Cases: Good for social media automation and simpler AI agent tasks.

Community: Relatively new, but expected to grow due to Python's popularity and support from AI16Z contributors.

GAME (VIRTUAL):

Focus: Autonomous, adaptive AI agents that evolve based on interactions in virtual environments.

Use Cases: Best suited for AI agents that learn and adapt to scenarios, such as games or virtual worlds.

Community: Innovative community, but still determining where to position itself against the competition.

3. Star data trend on Github

The above figure shows the GitHub star attention data since the release of these frameworks. It is worth noting that GitHub star is an indicator of community interest, project popularity and project perceived value.

ELIZA (red line):

The rise from a low base in July to a large increase in stars in late November (to 61,000 stars) indicates a rapid increase in interest and attracting the attention of developers. This exponential growth shows that ELIZA has gained significant traction due to its features, updates, and community engagement. Its popularity far exceeds that of other competitors, indicating that it has strong community support and broader applicability or interest in the AI ​​community.

RIG (blue line):

Rig is the oldest of the four frameworks and has seen a modest but consistent increase in stars, which is likely to increase significantly in the coming month. It has reached 1,700 stars but is still rising. Continued development, updates, and a growing number of users are the reasons for the continued accumulation of user interest. This may reflect that the framework has a niche user base or is still gaining reputation.

ZEREPY (yellow line):

ZerePy was just launched a few days ago and has accumulated 181 stars. It is worth emphasizing that ZerePy needs more development to increase its visibility and adoption. The cooperation with AI16Z may attract more code contributors.

GAME (green line):

This project has the least number of stars. It is worth noting that this framework can be directly applied to agents in the virtual ecosystem through API, thus eliminating the need for Github visibility. However, this framework was only publicly opened to builders more than a month ago, and more than 200 projects are being built using GAME.

4. Bullish reasons for the framework

The V2 version of Eliza will integrate the Coinbase proxy suite. All projects using Eliza will support native TEEs in the future, enabling agents to run in a secure environment. An upcoming feature of Eliza is the Plugin Registry, which will allow developers to seamlessly register and integrate plugins.

In addition, Eliza V2 will support automated anonymous cross-platform messaging. The token economics whitepaper is scheduled to be released on January 1, 2025, and is expected to have a positive impact on the underlying AI16Z token of the Eliza framework. AI16Z plans to continue to enhance the utility of the framework and continue to attract high-quality talent, and the efforts of its major contributors have proven that it has such ability.

The GAME framework provides code-free integration for agents, allowing GAME and ELIZA to be used simultaneously in a single project, each serving a specific purpose. This approach is expected to attract builders who focus on business logic rather than technical complexity. Although the framework has only been publicly released for 30 days, it has made substantial progress as the team works hard to attract more contributors. It is expected that all projects launched on VIRTUAL will use GAME.

Rig, represented by the ARC token, has great potential, although its framework is still in the early stages of growth and the plan to drive project adoption has only been online for a few days. But high-quality projects adopting ARC are expected to emerge soon, similar to the Virtual flywheel, but with the focus on Solana. The team is optimistic about the cooperation with Solana, comparing the relationship between ARC and Solana to Virtual to Base. It is worth noting that the team not only encourages new projects to start with Rig, but also encourages developers to enhance the Rig framework itself.

Zerepy is a newly launched framework that is gaining more and more attention due to its partnership with Eliza. The framework has attracted Eliza contributors who are actively improving it. Driven by ZEREBRO fans, it has a fervent following and provides new opportunities for Python developers, who were previously underrepresented in the competition for AI infrastructure. The framework will play an important role in AI creativity.