Deconstructing AI Framework: From Intelligent Agents to Decentralization Exploration

Introduction

The AI Agent track has developed rapidly recently, with increasing market attention. In just two months, the narrative combining AI and cryptocurrency has seen almost weekly changes. Recently, "framework" projects dominated by technological narratives have become the market focus, with several projects' market capitalizations surpassing hundreds of millions or even billions of dollars. These projects have given rise to new asset issuance models: issuing tokens based on GitHub repositories, and Agents built on frameworks can issue tokens again. Based on frameworks, with Agents as the application layer, a model similar to an asset issuance platform is being formed, which is essentially a unique infrastructure model of the AI era. This article will start with an introduction to frameworks and explore the impact of AI frameworks on the cryptocurrency field.

1. What is a framework?

The AI framework is a type of underlying development tool or platform that integrates pre-built modules, libraries, and tools, simplifying the construction process of complex AI models. The framework can be understood as the operating system of the AI era, similar to Windows and Linux in desktop systems, or iOS and Android in mobile devices. Each framework has its own advantages and disadvantages, allowing developers to choose based on their needs.

Although "AI frameworks" are an emerging concept in the cryptocurrency field, AI frameworks have a history of nearly 14 years since the birth of Theano in 2010. There are mature frameworks available in the traditional AI field, such as Google's TensorFlow and Meta's Pytorch.

The framework projects emerging in the cryptocurrency field are built upon the massive demand for Agents driven by the AI craze, and they have branched out into other tracks, forming AI frameworks in different subfields. Here are a few examples of mainstream frameworks:

1.1 Eliza

Eliza is a multi-Agent simulation framework for creating, deploying, and managing autonomous AI Agents. Developed in TypeScript, it has good compatibility and is easy to integrate with APIs.

Mainly aimed at social media scenarios, supporting multi-platform integration, such as Discord, X/Twitter, Telegram, etc. Supports functions such as PDF document analysis, link content extraction, audio and video processing, image analysis, and more.

Currently supports four types of use cases: AI assistant applications, social media roles, knowledge workers, and interactive roles.

Supported models include: open source model local inference ( such as Llama3, Qwen1.5), OpenAI API cloud inference, default configuration is Nous Hermes Llama 3.1B, and integrated with Claude to achieve complex queries.

1.2 G.A.M.E

G.A.M.E is an automated generation and management multimodal AI framework, primarily aimed at the design of intelligent NPCs in games. Its features allow low-code or even no-code users to participate in Agent design by simply modifying parameters.

The core design is a modular design that allows multiple subsystems to work together, including components such as the Agent prompt interface, perception subsystem, strategic planning engine, world context, and dialogue processing module.

Workflow: The developer starts the Agent through the Agent prompt interface, and the perception subsystem receives the input and passes it to the strategic planning engine. The strategic planning engine utilizes the memory system, world context, and information from the Agent library to formulate and execute action plans. The learning module continuously monitors the results of Agent actions and adjusts Agent behavior.

The application scenarios mainly focus on the decision-making, feedback, perception, and personality of agents in virtual environments, applicable to games and the metaverse.

1.3 Rig

Rig is an open-source tool written in Rust that simplifies the development of large language model ( LLM ) applications. It provides a unified operating interface for easy interaction with multiple LLM service providers and vector databases.

Core Features:

• Unified Interface
• Modular Architecture
• Type Safety
• High performance

Workflow: User requests pass through the provider abstraction layer, smart agents call tools or query vector storage to obtain information, and generate responses through retrieval-augmented generation mechanisms such as (RAG).

Application scenarios include question answering systems, document search tools, chatbots, virtual assistants, and content creation, etc.

1.4 ZerePy

ZerePy is an open-source framework based on Python that simplifies the process of deploying and managing AI Agents on the X platform. Inherited from the Zerebro project, it is more modular and easier to extend.

Provide command line interface (CLI) to manage AI Agent. The core architecture is based on modular design, including:

• LLM Integration: Supports OpenAI and Anthropic models
• X Platform Integration: Direct integration with X Platform API
• Modular Connection System: Convenient for adding support for other platforms or services
• Memory system ( planning ): Enables the Agent to remember previous interactions and contextual information.

Compared to Eliza, ZerePy focuses more on simplifying the process of deploying AI Agents on the X platform, leaning towards practical applications.

2. The Replica of the BTC Ecosystem

The development path of AI Agents is similar to that of the BTC ecosystem: GOAT/ACT - Social class Agents/Analytical class AI Agents - Framework competition. It is expected that infrastructure projects focusing on the decentralization and security of Agents will become the main theme in the next phase.

The AI framework project offers a new infrastructure development approach. Compared to the Memecoin Launchpad and the Inscription Protocol, the AI framework resembles a future public chain, while the Agent resembles a future Dapp.

Future debates may shift from EVM and heterogeneous chains to a framework rivalry. The key issue is how to achieve Decentralization or chainification, and the significance of developing AI frameworks on the blockchain.

3. What is the significance of going on-chain?

The core issue facing the combination of blockchain and AI is: Is it meaningful? Referring to the successful experiences of DeFi, the reasons supporting the Agent chain may include:

1. Reduce usage costs, improve accessibility and choice, allowing ordinary users to participate in AI "rental rights".

2. Provide blockchain-based security solutions to meet the needs of Agents interacting with real or virtual wallets.

3. Create unique blockchain financial gameplay, such as investment opportunities related to Agent's computing power, data tagging, etc.

4. Achieve a transparent and traceable reasoning process, improve interoperability, and be more attractive than the agent browsers provided by traditional internet giants.

4. Creative Economy

Framework projects may offer entrepreneurial opportunities similar to the GPT Store in the future. A framework that simplifies the agent building process and provides complex functionality combinations may have an advantage, creating a more interesting Web3 creative economy than the GPT Store.

Web3 has many unmet demands, and the economic system can make the policies of Web2 giants fairer. Introducing community economics helps improve the Agent. The creative economy of Agents will provide opportunities for ordinary people to participate, and future AI Memes may be smarter and more interesting than the Agents on existing platforms.