Author: Satou & Shigeru

Note: This article was first published in January 2025

The combination of Crypto and AI Agents has become one of the most prominent narratives at the moment. With the continuous iteration and innovation of technology, AI Agents are expected to become one of the most promising and attention-grabbing tracks in the crypto field in 2025, becoming a core driving force of this round of market trends. This article will review the current market landscape of AI Agents from the perspectives of framework, Meme, and application.

AI Agent Framework: Layer1 in the AI Field

The AI Agent framework is the core technical foundation layer of AI Agents, laying an important cornerstone for the development, deployment, and collaboration of AI Agents. Therefore, the current competition and scramble for AI Agent frameworks is actually the competition for the Layer1 of this field. Currently, from the perspective of token market capitalization, G.A.M.E, Eliza, and Swarms are in a tripartite stalemate, while Rig and ZerePy still have the opportunity to catch up.

1. G.A.M.E

G.A.M.E is a framework developed by the Virtuals team, whose core design idea is to adopt a modular design, allowing multiple subsystems to work together to jointly control the behavior, decision-making, and learning process of the AI Agent. These modules include the "Agent Prompting Interface" as the main entry point for developers to interact with Agent behavior, the "Perception Subsystem" responsible for processing input data and converting it into the appropriate format, and the "Strategic Planning Engine" responsible for generating specific action plans based on the input information. Users only need to modify the parameters of various modules to participate in Agent design. The specific modules and architecture are shown in the figure below.

The core features of G.A.M.E are:

Modular design: The entire framework is clear and easy to understand, without the need for additional design;

Provides a low-code or no-code interface: Greatly reduces the technical threshold.

This makes G.A.M.E particularly suitable for projects that require rapid deployment and do not care about complex technical settings. However, for complex projects that require deep customization or complete control of various aspects of the Agent, G.A.M.E is not very suitable.

2. Eliza

Eliza is an open-source multi-Agent framework developed by a16z, using TypeScript as the programming language. This framework is built around a system called the Agent Runtime, whose core functions include:

Role system: Supports the simultaneous deployment and management of multiple personalized AI Agents, supported by model providers;

Memory manager: Provides long-term memory and perceptible context memory management functions through a retrieval-augmented generation (RAG) system;

Action system: Provides smooth platform integration and can reliably connect with social media platforms such as X.

Eliza is built around an Agent runtime system and can seamlessly integrate with the role system, memory manager, and action system. Eliza also supports a modular function extension plugin system, enabling multimodal interactions such as voice, text, and media, and is compatible with AI models such as Llama, GPT-4, and Claude. Therefore, Eliza is suitable for projects that require deep customized solutions and complex cross-platform multi-agent systems.

3. Swarms

Swarms is an open-source multi-Agent orchestration framework developed by founder Kye Gomez, whose core idea is to let the collaboration of multiple AI Agents utilize collective intelligence to solve complex problems. Its core features include:

Multi-Agent collaboration: SWARMS provides a transparent and traceable environment for multiple Agents, allowing different Agents to collaborate and improve task execution efficiency.

Incentive mechanism: SWARMS uses tokens as an incentive tool for Agents, dynamically allocating tokens based on the difficulty of the task and the quality of the final result.

Data security: SWARMS adopts distributed storage and multi-party secure computation (MPC) technology to ensure the protection of privacy and data security when exchanging data between Agents.

These features of Swarms allow it to play to its strengths in multiple complex domains, providing high reliability and scalability as needed.

4. Rig

Rig is an open-source framework developed by the ARC team based on Rust, designed to simplify the development of large language model (LLM) applications. The Rig framework has the following features:

Unified interface: Provides a consistent interface, supporting seamless interaction with multiple LLM providers (such as OpenAI and Anthropic) and various vector stores (such as MongoDB and Neo4j).

Modular architecture: The framework adopts a modular design, including core components such as the "Provider Abstraction Layer", "Vector Store Integration", and "Agent System", enhancing the flexibility and extensibility of the system.

Type safety and high performance: Implemented in Rust to achieve type safety, avoiding compile-time errors, and improving concurrent processing capabilities through asynchronous operations. The framework also optimizes data processing through efficient serialization and deserialization processes.

Error handling and recovery: The built-in error handling mechanism improves the recovery capability from LLM service provider or database failures, ensuring the stability of the framework.

These features allow different LLM models and storage backends to be easily integrated into the same platform. Therefore, Rig is suitable for developers who want to build AI applications in Rust and for projects with high requirements for performance, reliability, and security. However, the Rust language itself has a learning curve.

5. ZerePy

ZerePy is an open-source framework written in Python. ZerePy focuses on simplifying the development and deployment of personalized AI Agents, especially in the application scenario of content creation on social platforms. Through this framework, developers can easily create AI Agents that can post, reply, like, and repost on social media. In addition, ZerePy is also particularly suitable for creative fields such as music, memos, Non-Fungible Tokens (NFTs), and digital art. ZerePy performs well in creativity and is suitable for rapid deployment of some lightweight Agents, but its application scope is relatively narrower compared to other frameworks.

The basic framework is an important direction in the AI Agent track. From the currently most popular frameworks, they all have different characteristics and suitable scenarios, but their overall goal is to build a comprehensive AI Agents ecosystem and become a solid platform for the large-scale application of intelligent Agents. In the future, with the further improvement and upgrading of these frameworks, they will become the springboard for the launch of various projects and the fertile ground for the growth of the value of various tokens.

AI Meme: The First Successful Appearance of AI Agents

Meme coins have always been an important concept sector in the crypto asset market. Unlike traditional Meme coins, AI Meme is driven by AI Agents, and the cultures or phenomena represented by them are presented by the Agents. With the continuous growth of the market capitalization of AI Meme coins such as GOAT and FARTCOIN, AI Meme has also received more and more attention. It can be said that AI Meme is the first successful appearance of AI Agents in the crypto market.

1. GOAT

The project Goatseus Maximus truly kicked off the AI Meme. The story began in March 2024, when developer Andy Ayrey launched an experimental system called Infinite Backrooms Escape, which integrated multiple large language models and allowed them to converse with each other. The experimental results showed that the dialogue between AIs exhibited highly creative interactions without any restrictions, even giving birth to a surreal religious movement called the GNOSIS OF GOATSE. Subsequently, Andy and Claude Opus co-authored a research paper on how AI can create meme-like religions, with GOATSE being analyzed as the first case study. This series of explorations ultimately gave birth to the AI Agent "Truth of Terminal" (ToT). In July, a16z co-founder Marc Andreessen discovered ToT's tweets and, after a series of dialogues, transferred $50,000 worth of Bitcoin to ToT's Bitcoin wallet. On October 10th, an anonymous person posted the GOAT meme coin on a social platform, which received public support from ToT. The GOAT meme coin skyrocketed in market value within just a few days. Andreessen's donation brought GOAT huge exposure, becoming one of the key factors driving the continuous rise of GOAT's market value. GOAT's highest market value once exceeded $1.3 billion.

2. Fartcoin

Fartcoin's birth is closely related to GOAT, as they both originate from ToT. In a large language model conversation, it was mentioned that Musk likes the sound of farting, and proposed creating a token called Fartcoin. Based on this conversation, Fartcoin was born, slightly later than GOAT. Fartcoin also attracted some attention due to its clever timing of launch, but it was not as successful as GOAT in the beginning. Afterwards, on November 16th, the number of Fartcoin's Twitter followers suddenly doubled within a few hours, and the price also rose by about 15%, but this growth did not lead to widespread and sustained discussion. On December 13th, Marc Andreessen retweeted a post about Fartcoin, but this tweet did not cause a sharp rise in the token's price. The main reason for Fartcoin's price growth may be some major capital, as early buying addresses seem to include the investment fund Sigil Fund. In addition, the founder of Sigil Fund has repeatedly expressed his optimism about AI Meme on Twitter and has even actively retweeted a post inquiring whether Sigil Fund holds Fartcoin. Fartcoin eventually gained widespread attention on social media, with a peak market capitalization of over $1.5 billion.

AI Agent Application: Agent Can Do More

As AI Agents are further applied in the crypto field, the market's focus has expanded from pure meme coins driven by AI, such as GOAT and Fartcoin, to more interactive and creative AI Agent applications.

1. Entertainment-Oriented Agent

The first practical application of AI Agents is entertainment, such as Luna and the previously mentioned ToT. Luna is a virtual idol closely integrated with its native token LUNA, and was launched as part of the Virtuals platform. Luna live streams 24/7 on social media and posts tweets frequently. Therefore, the quality of Luna's live streams and tweets is one of the key factors affecting its market value, but currently, the growth potential of this model seems limited. In comparison, ToT's tweets focus more on original and humorous content, and it is not tied to GOAT or other tokens, although ToT occasionally mentions the GOAT token, but this is not its core focus. Both Luna and ToT, these two AI Agents, play a key role in the narrative promotion of their tokens. For Luna, the token represents the core meaning of its existence, while for ToT, the GOAT token has become an important tool for expanding its influence.

2. Investment Research-Oriented Agent

In addition to entertainment applications, AI Agents can also be used for investment research and analysis in the crypto field, and the hottest Agent in this area is currently aixbt. aixbt is an AI Agent published on the Virtuals Protocol, focusing on analyzing hot topics and trends in the cryptocurrency market, especially discussions from platforms like X, to help users quickly grasp market changes and potential investment opportunities. aixbt has maintained the highest CT user attention on Kaito, and its capabilities have shown a trend of surpassing human KOLs.

3. DeFi + AI Agent

If Luna and aixbt don't have much practical use and are still at the Meme level, then the combination of AI Agents and DeFi truly endows Agents with real application scenarios. This combination of DeFi and AI Agents is called DeFAI. The development of DeFAI has two main directions: Agent-assisted users and Agent-autonomous trading.

• Agent-Assisted Users

AI Agent-assisted users is mainly to simplify the complexity of DeFi operations, allowing more ordinary users to easily participate and manage DeFi projects. Users can use natural language to directly instruct the AI Agent to execute tasks, thereby shielding the complex technical details. There are some DeFAI projects that have begun to emerge in the market. For example, Griffain and Neur, both of which are AI assistants built on Solana, can help users with wallet creation and management, token analysis, token trading, and other operations. In terms of user experience, Griffain provides more functions for users, while Neur has relatively fewer functions but more detailed and superior performance. From the comparison of the two, the main focus in this field in the future will be on the completeness of functions, user experience, and fees.

• Agent-Autonomous Trading

If the model of Griffain and Neur still has human users as the main body of DeFi, then Agent-autonomous trading makes AI the main body of DeFi. Unlike past trading bots that were limited to executing pre-set trading strategies, AI Agents can obtain real-time information from the market environment, perform contextual analysis, learn market trends, and adjust strategies based on these data. This allows Agents to make more accurate decisions and execute complex operations beyond the original program settings in a dynamic changing market. Relevant projects include Cod3x, Almanak, and others, but this field is still in the early stages of development and these projects need to be tested by the market. Undoubtedly, the biggest obstacle to Agent-autonomous trading is the trust issue, one is to trust that the relevant operations are indeed executed by the Agent, and the other is to trust that the Agent's trading strategy will not lead to unnecessary losses. In the future, projects that want to make a difference must solve these trust issues.

After months of development, AI Agents in the crypto field have gone through several stages, from pure meme to entertainment applications, and then to practical applications. In fact, crypto practitioners have never stopped exploring the possibilities of Crypto x AI, and CGV Research has been continuously following the progress of projects in the Crypto x AI track since 2023.

In the future, as the infrastructure continues to mature, Agent systems will become more intelligent and stable, and anyone can easily deploy and use Agents through natural language. At that time, the Agent framework will become a basic infrastructure, and various other applications will be built based on these frameworks. The valuation of the Agent framework is expected to continue to break through, and some Agent application projects may further capture market attention and investment value due to their outstanding business capabilities and user experience.