AI Agent：Web3智慧輔助新時代

今年下半年以來，AI Agent的話題持續升溫。起初，AI聊天機器人終端的真相吸引了廣泛關注

With the continuous heating up of the AI Agent topic in the second half of this year, at first, the truths AI chatbot terminal attracted widespread attention for its humorous posts and replies on X (similar to "Robert" on Weibo), and received a $50,000 grant from a16z founder Marc Andreessen. Inspired by its published content, someone created the GOAT token, which rose by more than 10,000% in just 24 hours. The topic of AI Agent immediately attracted the attention of the Web3 community. Later, the first decentralized AI trading fund based on Solana, ai16z, came out, launched the AI Agent development framework Eliza, and triggered a dispute over uppercase and lowercase tokens. However, the community still has an unclear concept of AI Agent: What is the core of AI Agent? How is it different from the Telegram trading robot?

隨著今年下半年AI代理話題的持續升溫，起初，真相AI聊天機器人終端因其在X上幽默的發帖和回复（類似微博上的“羅伯特”）而引起廣泛關注，並獲得了廣泛關注。 a16z 創辦人Marc Andreessen 提供50,000 美元資助。受到其發佈內容的啟發，有人創建了 GOAT 代幣，該代幣在短短 24 小時內上漲了 10,000% 以上。 AI Agent的話題立刻引起了Web3社群的注意。後來，第一個基於Solana的去中心化AI交易基金ai16z問世，推出了AI Agent開發框架Eliza，並引發了大小寫代幣之爭。然而，業界對於AI Agent的概念仍然不明確：AI Agent的核心是什麼？它與 Telegram 交易機器人有何不同？

How it works: Perception, reasoning, and autonomous decision-making

運作原理：感知、推理與自主決策

AI Agent is an intelligent agent system based on a large language model (LLM) that can perceive the environment, make reasoning decisions, and complete complex tasks by calling tools or performing operations. Workflow: Perception module (obtaining input) → LLM (understanding, reasoning and planning) → tool calling (task execution) → feedback and optimization (verification and adjustment).

AI Agent是基於大語言模型（LLM）的智慧代理系統，可以透過呼叫工具或執行操作來感知環境、做出推理決策並完成複雜的任務。工作流程：感知模組（獲取輸入）→LLM（理解、推理和規劃）→工具呼叫（任務執行）→回饋和最佳化（驗證和調整）。

Specifically, AI Agent first obtains data (such as text, audio, images, etc.) from the external environment through the perception module and converts it into structured information that can be processed. As a core component, LLM provides powerful natural language understanding and generation capabilities, acting as the "brain" of the system. Based on the input data and existing knowledge, LLM performs logical reasoning to generate possible solutions or formulate action plans. Subsequently, AI Agent completes specific tasks by calling external tools, plug-ins or APIs, and verifies and adjusts the results based on feedback to form a closed-loop optimization.

具體來說，AI Agent首先透過感知模組從外部環境獲取資料（如文字、音訊、圖像等），並將其轉換為可處理的結構化資訊。 LLM作為核心組件，提供強大的自然語言理解和生成能力，充當系統的「大腦」。基於輸入資料和現有知識，LLM進行邏輯推理以產生可能的解決方案或製定行動計劃。隨後，AI Agent透過呼叫外部工具、插件或API完成特定任務，並根據回饋對結果進行驗證和調整，形成閉環優化。

In the application scenarios of Web3, what is the difference between AI Agent and Telegram trading robots or automated scripts? Take arbitrage as an example. Users want to conduct arbitrage transactions under the condition that the profit is greater than 1%. In Telegram trading robots that support arbitrage, users set up trading strategies with profits greater than 1%, and the Bot begins to execute. However, when the market fluctuates frequently and arbitrage opportunities are constantly changing, these Bots lack risk assessment capabilities and execute arbitrage as long as the profit is greater than 1%. In contrast, AI Agent can automatically adjust its strategy. For example, when the profit of a transaction exceeds 1%, but through data analysis, the risk is too high, and the market may suddenly change and cause losses, it will decide not to execute the arbitrage.

在Web3的應用場景中，AI Agent與Telegram交易機器人或自動化腳本有何不同？以套利為例。用戶希望在利潤大於1%的條件下進行套利交易。在支援套利的Telegram交易機器人中，用戶設定利潤大於1%的交易策略，Bot開始執行。但當市場波動頻繁、套利機會不斷變化時，這些Bot缺乏風險評估能力，只要利潤大於1%就會執行套利。相比之下，AI Agent可以自動調整策略。例如，當一筆交易的利潤超過1%，但透過數據分析，風險太大，市場可能突然發生變化而造成損失時，就會決定不執行套利。

Therefore, AI Agent is self-adaptive. Its core advantage lies in its ability to self-learn and make decisions autonomously. It can adjust its behavior strategy based on feedback signals through interaction with the environment (such as the market, user behavior, etc.) to continuously improve the performance of task execution. It can also make decisions in real time based on external data and continuously optimize its decision-making strategy through reinforcement learning.

因此，AI Agent具有自適應性。其核心優勢在於具有自學習、自主決策的能力。它可以透過與環境（如市場、使用者行為等）的交互，根據回饋訊號調整自身的行為策略，不斷提高任務執行的效能。它還可以根據外部數據即時做出決策，並透過強化學習不斷優化其決策策略。

Does this sound a bit like a solver in the intent framework? AI Agent itself is also a product based on intent. The biggest difference between it and the solver in the intent framework is that the solver relies on precise algorithms and is mathematically rigorous, while AI Agent decision-making relies on data training, and often requires continuous trial and error during the training process to approach the optimal solution.

這聽起來是不是有點像是意圖框架中的解算者？ AI Agent本身也是一個基於意圖的產品。它與意圖框架中的求解器最大的區別在於，求解器依賴精確的演算法，在數學上是嚴謹的，而AI Agent的決策則依賴資料訓練，往往需要在訓練過程中不斷試誤才能逼近目標。

AI Agent Mainstream Framework

AI Agent主流框架

AI Agent framework is an infrastructure for creating and managing intelligent agents. Currently in Web3, popular frameworks include Eliza from ai16z, ZerePy from zerebro, and GAME from Virtuals.

AI Agent框架是用於建立和管理智慧代理的基礎架構。目前在Web3中，流行的框架包括ai16z的Eliza、zerebro的ZerePy和Virtuals的GAME。

Eliza is a versatile AI Agent framework built with TypeScript. It supports running on multiple platforms (such as Discord, Twitter, Telegram, etc.), and through complex memory management, it can remember previous conversations and contexts, and maintain stable and consistent personality traits and knowledge answers. Eliza uses the RAG (Retrieval Augmented Generation) system, which can access external databases or resources to generate more accurate answers. In addition, Eliza integrates a TEE plug-in, allowing deployment in TEE to ensure data security and privacy.

Eliza 是一個使用 TypeScript 建構的多功能 AI 代理框架。它支援在多個平台上運行（如Discord、Twitter、Telegram等），並透過複雜的記憶體管理，可以記住先前的對話和上下文，並保持穩定一致的人格特徵和知識答案。 Eliza 使用 RAG（檢索增強生成）系統，該系統可以存取外部資料庫或資源以產生更準確的答案。此外，Eliza整合了TEE插件，允許部署在TEE中以確保資料安全和隱私。

GAME is a framework that enables and drives AI Agents to make autonomous decisions and actions. Developers can customize the behavior of agents according to their needs, expand their functions, and provide customized operations (such as social media posting, replying, etc.). Different functions in the framework, such as the agent's environmental location and tasks, are divided into multiple modules to facilitate developers to configure and manage. The GAME framework divides the decision-making process of AI Agents into two levels: high-level planning (HLP) and low-level planning (LLP), which are responsible for tasks and decisions at different levels. High-level planning is responsible for setting the overall goals and task planning of the agent, making decisions based on goals, personality, background information and

GAME 是一個框架，可支援並驅動 AI 代理程式做出自主決策和行動。開發者可依需求客製化代理商的行為，擴展其功能，提供客製化操作（如社群媒體貼文、回覆等）。框架中的不同功能，如代理的環境位置、任務等，被劃分為多個模組，方便開發者配置和管理。 GAME架構將AI Agent的決策過程分為兩個層次：高階規劃（HLP）和低階規劃（LLP），分別負責不同層級的任務和決策。高層規劃負責設定智能體的總體目標和任務規劃，根據目標、個性、背景資訊和