Introduction

In previous articles, we have frequently discussed our views on the current state of AI Memes and the future development of AI Agents. However, the rapid narrative development and evolution of the AI Agent track have still been somewhat overwhelming. In the brief two months since the launch of “Truth Terminal” and the beginning of Agent Summer, the narrative of AI and Crypto integration has evolved almost weekly. Recently, market attention has started to shift towards “framework” projects, which are primarily driven by technological narratives. This niche subfield has already produced several unicorn projects with a market cap of over a billion dollars within the past few weeks. These projects have also led to a new paradigm for asset issuance, where projects issue tokens based on their GitHub code repositories, and the Agents built on these frameworks can also issue tokens. At the core of this structure, we have frameworks, with Agents as the layer above. It resembles an asset issuance platform, yet it is actually a unique infrastructural model emerging in the AI era. How should we view this new trend? This article will start with an introduction to frameworks and offer an interpretation of what AI frameworks mean for Crypto, combining these insights with our own reflections.

I. What is a Framework?

By definition, an AI framework is an underlying development tool or platform that integrates a set of pre-built modules, libraries, and tools to simplify the process of building complex AI models. These frameworks typically also include functions for handling data, training models, and making predictions. In simple terms, you can think of a framework as an operating system for the AI era, similar to desktop operating systems like Windows or Linux, or mobile operating systems like iOS and Android. Each framework has its own advantages and disadvantages, allowing developers to choose based on their specific needs.

Although the term “AI framework” is still a relatively new concept in the Crypto field, its development actually dates back nearly 14 years, starting with Theano in 2010. In the traditional AI community, both academia and industry have already developed very mature frameworks to choose from, such as Google’s TensorFlow, Meta’s PyTorch, Baidu’s PaddlePaddle, and ByteDance’s MagicAnimate, each of which has its advantages in different scenarios.

The AI framework projects currently emerging in Crypto are based on the demand for a large number of Agents arising from the AI boom, and these have further branched out into other tracks in Crypto, eventually forming different AI frameworks for specific subfields. Let’s explore some of the current mainstream frameworks in the industry to better illustrate this point.

1.1 Eliza

First, let’s consider Eliza, a framework created by ai16z. It is a multi-agent simulation framework designed to create, deploy, and manage autonomous AI agents. Developed using TypeScript as the programming language, its advantage lies in better compatibility and easier API integration. According to official documentation, Eliza is primarily designed for social media, offering support for multi-platform integrations. The framework provides full-featured Discord integration, supporting voice channels, automated accounts for X/Twitter, Telegram integration, and direct API access. In terms of media content processing, it supports reading and analyzing PDF documents, extracting and summarizing links, audio transcription, video content handling, image analysis, and conversation summaries.

The use cases supported by Eliza currently include the following four categories:

1. AI Assistant Applications: Customer support agents, community administrators, personal assistants.
2. Social Media Roles: Automated content creators, interactive bots, brand representatives.
3. Knowledge Workers: Research assistants, content analysts, document processors.
4. Interactive Roles: Role-playing characters, educational tutors, entertainment bots.

The models currently supported by Eliza are:

1. Open-source local inference models: such as Llama3, Qwen1.5, BERT.
2. Cloud inference via OpenAI API.
3. Default configuration as Nous Hermes Llama 3.1B.
4. Integration with Claude for complex queries.

1.2 G.A.M.E

G.A.M.E (Generative Autonomous Multimodal Entities Framework) is a multimodal AI framework for automatic generation and management, launched by Virtual. It is primarily designed for intelligent NPC design in games. One unique aspect of this framework is that it allows even low-code or no-code users to participate in Agent design by simply modifying parameters through its trial interface.

In terms of project architecture, G.A.M.E is built on a modular design, where multiple subsystems work together in collaboration. The detailed architecture is as follows:

1. Agent Prompting Interface: The interface for developers to interact with the AI framework. Through this interface, developers can initiate a session and specify session IDs, agent IDs, user IDs, and other parameters.
2. Perception Subsystem: Responsible for receiving input information, synthesizing it, and sending it to the strategic planning engine. It also handles responses from the dialogue processing module.
3. Strategic Planning Engine: The core of the entire framework, divided into a high-level planner and low-level policy. The high-level planner is responsible for formulating long-term goals and plans, while the low-level policy translates these plans into specific actions.
4. World Context: Contains environmental information, world status, and game state data, helping agents understand their current context.
5. Dialogue Processing Module: Handles messages and responses, generating dialogue or reactions as output.
6. On-Chain Wallet Operator: Likely related to blockchain technology applications, though specific functions are unclear.
7. Learning Module: Learns from feedback and updates the agent’s knowledge base.
8. Working Memory: Stores recent actions, results, and current plans, among other short-term information.
9. Long-Term Memory Processor: Extracts and ranks important information about the agent and its working memory based on factors like importance, recency, and relevance.
10. Agent Repository: Stores the agent’s goals, reflections, experiences, and characteristics.
11. Action Planner: Generates specific action plans based on low-level strategies.
12. Plan Executor: Executes the action plans generated by the action planner.

Workflow: Developers initiate an Agent via the Agent Prompting Interface, where the Perception Subsystem receives input and sends it to the Strategic Planning Engine. The engine, with the help of the memory system, world context, and Agent repository, formulates and executes an action plan. The Learning Module monitors the Agent’s actions and adjusts its behavior accordingly.

Application Scenarios: From the overall technical architecture, this framework focuses on decision-making, feedback, perception, and personality of Agents in virtual environments. In addition to gaming, this framework is also applicable to the Metaverse. The list below from Virtual shows that many projects have already adopted this framework for construction.

1.3 Rig

Rig is an open-source tool written in Rust, specifically designed to simplify the development of large language model (LLM) applications. It provides a unified interface that allows developers to easily interact with multiple LLM service providers (such as OpenAI and Anthropic) and various vector databases (like MongoDB and Neo4j).

Key Features:

• Unified Interface: Regardless of which LLM provider or vector storage is used, Rig offers a consistent access method, greatly reducing the complexity of integration work.
• Modular Architecture: The framework uses a modular design, which includes key components such as “Provider Abstraction Layer,” “Vector Storage Interface,” and “Intelligent Agent System,” ensuring flexibility and scalability of the system.
• Type Safety: Leveraging Rust’s features, Rig achieves type-safe embedding operations, ensuring code quality and runtime security.
• High Performance: The system supports asynchronous programming, optimizing concurrency processing capabilities. Built-in logging and monitoring features help with maintenance and troubleshooting.

Workflow: When a user enters the Rig system, the request first goes through the “Provider Abstraction Layer,” which standardizes the differences between various providers and ensures consistent error handling. In the core layer, intelligent agents can call different tools or query vector storage to retrieve the necessary information. Finally, advanced mechanisms such as Retrieval-Augmented Generation (RAG) combine document retrieval and contextual understanding to generate accurate and meaningful responses before returning them to the user.

Use Cases: Rig is suitable for building systems that require fast and accurate question answering, creating efficient document search tools, developing context-aware chatbots or virtual assistants, and even supporting content creation by automatically generating text or other forms of content based on existing data patterns.

1.4 ZerePy

ZerePy is an open-source framework based on Python designed to simplify the deployment and management of AI agents on the X (formerly Twitter) platform. It evolved from the Zerebro project and inherited its core functionalities but was designed in a more modular and scalable manner. The goal is to allow developers to easily create personalized AI agents and implement various automation tasks and content creation on X.

ZerePy provides a command-line interface (CLI), making it convenient for users to manage and control the AI agents they deploy. Its core architecture is modular, allowing developers to flexibly integrate different functional modules, such as:

• LLM Integration: ZerePy supports large language models (LLMs) from OpenAI and Anthropic, allowing developers to select the model best suited to their application. This enables agents to generate high-quality textual content.
• X Platform Integration: The framework directly integrates with X’s API, enabling agents to perform tasks such as posting, replying, liking, and retweeting.
• Modular Connection System: This system allows developers to easily add support for other social platforms or services, extending the framework’s functionality.
• Memory System (Future Plans): Although not fully implemented in the current version, ZerePy’s design goal includes integrating a memory system that would allow agents to remember previous interactions and contextual information to generate more coherent and personalized content.

While both ZerePy and a16z’s Eliza project aim to build and manage AI agents, they differ in architecture and focus. Eliza is more oriented toward multi-agent simulations and broader AI research, while ZerePy focuses on simplifying the deployment of AI agents on specific social platforms (X), making it more application-oriented.

II. A Replica of the BTC Ecosystem

In terms of development path, AI agents share many similarities with the BTC ecosystem from late 2023 to early 2024. The development trajectory of the BTC ecosystem can be simply summarized as: BRC20-Atomical/Rune and other multi-protocol competition — BTC L2 — BTCFi centered around Babylon. While AI agents have developed more rapidly on the foundation of mature traditional AI technology stacks, their overall development path mirrors the BTC ecosystem in several respects. I would summarize it as follows: GOAT/ACT — Social-type agents — Analytical AI agent framework competition. From a trend perspective, infrastructure projects focused on decentralization and security around agents will likely also carry this framework wave forward, becoming the next dominant theme.

So, will this track, like the BTC ecosystem, lead to homogenization and bubbleization? I don’t think so. First, the narrative of AI agents is not about recreating the history of smart contract chains. Second, whether these existing AI framework projects are technically strong or still stuck in the PPT stage or merely Ctrl+C and Ctrl+V, at least they provide a new infrastructure development approach. Many articles have compared AI frameworks to asset issuance platforms, and agents to assets. However, compared to Memecoin Launchpads and Inscription protocols, I personally believe AI frameworks resemble future public chains, while agents resemble future DApps.

In today’s Crypto space, we have thousands of public chains and tens of thousands of DApps. In the realm of general-purpose chains, we have BTC, Ethereum, and various heterogeneous chains, while the forms of application chains are more diverse, such as game chains, storage chains, and Dex chains. Public chains and AI frameworks are quite similar in nature, and DApps can correspond well to agents.

In the era of Crypto in AI, it is highly likely that the space will evolve in this direction, with future debates shifting from EVM versus heterogeneous chains to framework debates. The current issue is more about decentralization, or how to “chainify” it. I believe that future AI infrastructure projects will develop around this foundation. Another important point is: What significance does doing this on the blockchain hold?

III. The Significance of On-Chain

Regardless of what blockchain combines with, it ultimately faces one critical question: Is it meaningful? In last year’s article, I criticized GameFi for its misplaced priorities, where infrastructure development was overly advanced, and in previous articles about AI, I expressed skepticism about the current practicality of combining AI with Crypto. After all, the narrative-driving force for traditional projects has increasingly weakened. The few traditional projects that performed well last year in terms of token price were generally those that could match or exceed the price strength.

What can AI do for Crypto? Previously, I thought of use cases like AI agents performing tasks on behalf of users, Metaverse, and agents as employees — relatively mundane ideas but with certain demands. However, these demands do not require being fully on-chain, and from a business logic perspective, they cannot form a closed-loop. The agent browser mentioned in the last article, which implements intentions, could generate demands for data labeling and inference computing power, but these two elements are still not tightly integrated, and in terms of computational power, centralized computing still holds the advantage.

Revisiting the success of DeFi, the reason DeFi managed to carve out a slice of traditional finance is because it provides greater accessibility, better efficiency, lower costs, and trustless security. If we consider this framework, I think there may be several reasons why agent “chainization” might make sense:

1. Cost Reduction: Can agent chainization lower usage costs, thereby achieving greater accessibility and more choices for users? This could eventually allow ordinary users to participate in what has traditionally been the exclusive domain of Web2 tech giants’ AI “rentals.”
2. Security: According to the simplest definition, an agent is an AI that can interact with the virtual or real world. If an agent can intervene in the real world or even in my virtual wallet, then security solutions based on blockchain could become a necessity.
3. Blockchain-Specific Financial Play: Can agents create a unique set of financial mechanisms on the blockchain? For example, in AMM (Automated Market Maker), liquidity providers (LPs) allow ordinary users to participate in automated market-making. Similarly, if agents require computing power or data labeling, users could invest in these protocols in the form of USDT, based on their confidence in the system. Or, agents in different application scenarios could form new financial structures.
4. DeFi Interoperability: While DeFi currently lacks perfect interoperability, agents might be able to address this issue by enabling transparent, traceable reasoning processes, filling the gaps.

IV. Creativity?

Framework projects in the future will also provide entrepreneurial opportunities similar to the GPT Store. While launching an agent via a framework is still complex for ordinary users, I believe simplifying the agent construction process and providing more complex function combinations will give such frameworks a competitive advantage in the future. This could lead to the creation of a Web3 creative economy that is far more interesting than the GPT Store.

At present, the GPT Store is still more oriented toward traditional practical uses, with most of the popular apps being created by traditional Web2 companies. Moreover, the income generated is largely monopolized by the creators. According to OpenAI’s official explanation, the strategy is simply to provide financial support to outstanding developers in the United States, offering subsidies up to a certain amount.

From a demand perspective, Web3 still has many gaps to fill, and from an economic system standpoint, it can make Web2 giants’ unfair policies more equitable. Additionally, we can naturally introduce community economies to further improve agents. The creative economy around agents will present ordinary people with an opportunity to participate. In the future, AI memes will be far smarter and more interesting than the agents issued by GOAT or Clanker.

About YBB

YBB is a web3 fund dedicating itself to identify Web3-defining projects with a vision to create a better online habitat for all internet residents. Founded by a group of blockchain believers who have been actively participated in this industry since 2013, YBB is always willing to help early-stage projects to evolve from 0 to 1.We value innovation, self-driven passion, and user-oriented products while recognizing the potential of cryptos and blockchain applications.

Website | Twi: @YBBCapital

Disclaimer:

1. This article is reprinted from [YBB Capital]. All copyrights belong to the original author [YBB Capital Researcher Zeke]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute investment advice.
3. The Gate Learn team translated the article into other languages. Copying, distributing, or plagiarizing the translated articles is prohibited unless mentioned.