Author: Jinming Source: HashKey Capital Translation: Shan Oppa, Golden Finance

Introduction

The concept of artificial intelligence agents (AI Agent) refers to an intelligent software system that can understand its environment and perform actions independently on behalf of users or machines to achieve their goals. It was proposed as early as the 1980s. However, this concept started to gain attention until the 2010s, with the rise of deep learning and large language models (LLMs), demonstrating their ability to understand and generate human-like responses.

Today, LLM has become an integral part of our lives, and products like ChatGPT have more than 15.5 million paid users around the world, and demand will surely grow further as OpenAI launches smarter inference models. The widespread use of LLMs such as ChatGPT, Claude and DeepSeek paves the way for the natural development of the agency economy. Agent is more complex than LLM, and it is defined as a system consisting of a single model or multiple models, and a framework with a tool set and defining the identity of the proxy (Figure 1).

Agents equipped with roles and toolkits can receive tasks, analyze, process and perform actions independently on behalf of users, although sometimes human involvement is required to provide feedback and learn through reinforcement learning. Agents are essentially composable, and as agents become more specialized and technologically mature, the manual participation part of the agent system may take a back seat, and communication between agents will become the focus of simplifying complex workflows and releasing efficiency improvements. As agent-based frameworks continue to advance, we expect exponential benefits in a variety of applications through integrated blockchain, a technology based on transparency, decentralization and incentive consistency.

In addition, by leveraging the trustworthy, secure and transparent nature of blockchain technology, agents on smart contracts can perform autonomous wallet transactions, receive token incentives for good behavior, and be punished for confrontational behavior. In this report, we will first explore what multi-agent systems are and the orchestration frameworks that support the development of these systems, and then understand the synergy between multi-agent systems and Web3 technology. We will then explore the use cases, challenges, and problem-solving efforts of the Web3 multi-agent framework.

Figure 1: Components of the proxy

Multi-agent system

In multi-agent system, unlike single-agent systems, agents can focus on their respective fields and collaborate to simulate human teamwork and effectively solve multi-step, complex real-world problems (Figure 2). This enhances the cognitive and reasoning capabilities of agents based on individual LLMs, providing greater scalability and efficiency. In agents based on individual LLMs, agents bear the arduous burden of completing tasks from beginning to end, which often leads to delays and bottlenecks when tasks become more complex and demanding.

In multiagent systems, there is usually a task manager that defines task requirements, breaks tasks into smaller tasks, and delegates subtasks to agents based on their capabilities, thus making the multiagent system more resilient and suitable for large-scale enterprise use cases. The collaborative nature of multiagent systems facilitates efficient memory management by allowing each agent to store only the context related to its role. Thanks to its distributed architecture, agents avoid handling huge memory loads, thereby improving scalability and opening the door to a wider range of use cases.

The key to the development of multi-agent systems lies in multi-agent frameworks, which enable agents to communicate and coordinate effectively with each other to achieve established goals. Through a variety of multi-agent frameworks, multi-agent reinforcement learning (MARL), simulated environments and improved agent orchestration layers, they open up exciting opportunities for agent-driven applications across industries, including crypto industries. Below, we will look at some multiagent orchestration frameworks in Web2 and Web3 that unlock new possibilities through agent-driven workflows.

Multi-agent orchestration framework

Multi-agent orchestration framework handles the management of LLM-based agents to solve problems. Compared with a single agent, multi-agent systems play an important role in simplifying and improving efficiency when automating complex tasks.

Figure 2: Multi-agent framework architecture

Note that because the multi-agent framework continues to evolve, this is not an exhaustive list.

AutoGen

AutoGen is an open source multi-agent framework designed by the cutting-edge laboratory of artificial intelligence at Microsoft Research Institute. It facilitates the development of multi-agent applications with its modular and scalable design. AutoGen Core implements messaging and event-driven agents that can be programmed in Python and .NET languages. The AgentChat API enables seamless communication between agents and is built on top of the Core API. Various extensions are available to enable the agent to perform various functions such as web browsing, video analysis, file analysis, and encapsulating Langchain tools. MagenticOne, built on the AutoGen multi-agent framework, can perform tasks such as code, browse web pages, and manage files.

CrewAI

CrewAI

CrewAI is an open source multi-agent platform that enables efficient and seamless task automation through well-defined role-based multi-agent orchestration. Its architecture allows agents with configurable roles, goals, and personalities to interact sequentially or in parallel, ensuring orderly task execution. To maintain relevance, agents can leverage an extensive knowledge base that supports text sources and structured data formats. CrewAI also provides access to LangChain and LlamaIndex tools, as well as enterprise-level features provided by Portkey, enabling agents to easily use external APIs, databases, and retrieval systems. The platform is also developer-friendly and supports YAML-based configurations, which allows developers to easily configure and deploy agents.

Langroid

Langroid

Langroid is an open source Python programming framework that takes multi-agent programming as its core design principle, giving agents a similar status to citizens. Recognized by developers for its simplicity, intuitiveness and scalability, the framework provides a variety of modules and tools to meet the needs of complex agent applications. By default, agents act as message converters and have 3 responder methods: LLM responders, agent responders, and user responders. Together, these responder methods allow agents to perform functions, generate human-readable natural language responses, and incorporate human feedback into their agent workflow. Encapsulation of tasks around agents allows them to orchestrate interactions by delegating subtasks to other agents. The ToolMessage mechanism supports OpenAI LLM and LLM function calls, and the agent can access various tools and functions. Combined with integration with vector storage such as LanceDB, Qdrant and Chroma, Langroid's agent hasPersistent dialogue state and vector storage memory make it good at managing complex dynamic scenarios.

CAMEL

CAMEL

CAMEL is an open source multi-agent framework that provides a general infrastructure for a wide range of applications such as task automation, data generation and real-world simulation. As part of CAMEL, social modules play a crucial role in multiagent coordination. It contains two frameworks—RolePlaying and BabyAGI—that aims to manage agent interactions and drive goal-oriented outcomes. Its role-playing, dialogue-oriented approach makes it very suitable for building customer-facing agents. CAMEL's integration with a variety of vector databases and LLMs supports RAG and provides its agents with persistent memory, making it ideal for large-scale enterprise applications. However, the success of the RolePlaying framework currently requires developers to have effective tip engineering skills and character design, which may make it less friendly to those without a strong coding and AI background. CAMEL has deployed an AI chatbot, Eigent Bot, which can obtain real-time information, support multimodal functionality, and utilize graphical RAG for better context understanding.

MetaGPT

MetaGPT is a meta-programmed multi-agent orchestration framework that encodes standard operating procedures (SOPs) into prompt sequences, combining well-defined agent roles and responsibilities. This design helps mitigate the more complex hallucinatory risks posed by interagent interactions. Agents in MetaGPT communicate through defined output formats to a shared message pool instead of one-to-one conversations, reducing irrelevant or lost content. It also implements an executable feedback mechanism that supports self-correction and review. MetaGPT is especially effective in software development environments where well-defined roles can improve code quality and task allocation. MetaGPT achieved significant results in HumanEval and MBPP, at 85.9% and 87.7%, respectively, when measured based on code generation benchmarks.

LangGraph

LangGraph is an open source agent framework developed by the creator of LangChain. It aims to manage complex multi-agent workflows with a modular architecture that enables different agents to communicate, coordinate and execute tasks efficientlyServing. By using graph-based architecture to model the relationship between different components of an agent's workflow, LangGraph promotes dynamic task allocation, seamless scalability, and powerful problem-solving capabilities across distributed systems. This innovative approach simplifies state management and is suitable for multi-step workflows that require persistent context retention. Additionally, the Langchain Model Context Protocol (MCP) adapter (a lightweight wrapper) allows MCP tools to be easily converted into Langchain tools for use by LangGraph agents, thus extending their available toolsets. In the multiagent field, LangGraph benefits from a strong network effect as it leverages the LangChain ecosystem.

ElizaOS

ElizaOS

ElizaOS is perhaps the most famous Web3 multi-agent framework, an open source TypeScript multi-agent framework that embeds Web3 components to solve the barriers to entry and accessibility in the encryption industry. The framework is designed with a modular design with an extensive set of plug-ins and is currently able to support a range of models (i.e. OpenAI, DeepSeek, Llama, Qwen, etc.), platform integration (i.e. Twitter, Discord, Telegram, Farcaster, etc.), and over 25 chain compatibility (i.e. Solana, Ethereum, Ton, Aptos, Sui, Sei, etc.). Its integration with the GOAT SDK also enables agents to perform various on-chain operations. The core architecture of ElizaOS consists of agents, role files, providers, operations and evaluators, which together enable agents to have persistent memory and context-awareness when performing various tasks, and get feedback from the evaluators to ensure better performance.

A noteworthy example is the ai16z DAO Fund, which uses the ElizaOS framework to create an autonomous agent that can filter market signals and trade various meme coins. At its heyday, it managed over $36 million in AUM.

As the most mature agent framework in Web3, the ElizaOS agent framework continues to be popular with Web3 developers as it gains over 14K github stars and currently has 99 integrations. With the future planning to launch an agent startup platform, this could be done by providing developers with a no-need/low-code agent startup platform.Further stimulate their interest.

RIG

Another popular Web3 agent framework with more than 3K github stars is RIG, a Rust-based open source agent framework that stands out by providing lightweight cores while supporting advanced inference patterns (from prompt chains to conditional logic and parallel task execution). The RIG framework provides a unified API between supported LLM providers (OpenAI, cohere, DeepSeek, etc.) and provides simplified embedding and vector storage support for RAG implementations. You can also create custom tools that make the framework extensible for LLM-based applications.

Utilizing Rust's asynchronous function, multi-agent systems can handle multiple tasks concurrently. Although it currently lags behind ElizaOS in 23 Web3 native integrations. The developer behind RIG, ARC, has partnered with the Solana Foundation to drive adoption of the framework by providing targeted grants to developers who use RIG to build Rust-based agents. In addition, ARC has launched its agent startup platform Forge, which adopts a similar startup platform model to Virtuals, but currently only allows whitelisted teams to access the platform. A notable use case for the RIG and Forge launch platforms is the AskJimmy platform, a multiagent hedge fund that coordinates a group of agents powered by a trading strategy library to seamlessly execute transactions across EVM and Solana across leading platforms such as Hyperliquid, Drift, GMX.

G.A.M.E

G.A.M.E

The G.A.M.E framework developed by the Virtuals Protocol team is an open source multi-agent framework based on Python and JavaScript, which promotes the creation of on-chain agents. Its integration with the Web3 library GOAT SDK provides agents with over 200 on-chain operations across various protocols. Task processing is done through a hierarchical approach, where the task planner breaks the task into subtasks and delegates it to a professional working agent that coordinates and communicates to deliver the final output. Currently, most of its agents revolve around social media platforms and in-game environments, the most famous agent is AIXBT. Since its launch, AIXBT (a one with its own X accountAI-driven on-chain analytics influencer) is widely recognized for his analytical insights and has more than 490,000 followers as of this writing.

Source: Virtuals Protocol GAME architecture

uAgents

uAgents

uAgents is a Python-based multi-agent framework developed by Fetch.AI. It has been integrated with various Web2 frameworks such as LangChain, Vertex AI, CrewAI, etc., so that autonomous agents can be easily created and deployed on the Fetch.AI blockchain. Once created, the agent will be registered on the Almanac smart contract, allowing other agents to easily query the contract and identify the recipient agent through their agent address and HTTP endpoint. Encryption security ensures that interactions between agents remain secure, allowing the most appropriate agent to satisfy user requests without compromising security.

Comparative Analysis (Web2 Framework and Web3 Framework)

Advantages of Web3 Multi-Agent Framework

Environmental Framework

Although Web2 Multi-Agent Framework has been relatively mature and has gained strong institutional demand, they lack native on-chain capabilities compared to Web3 Multi-Agent Framework. Developers using Web2 tools must attach a third-party library to interact with smart contracts or parse blockchain data, thereby introducing complexity and potential vulnerabilities. Developers using Web3 multi-agent frameworks can benefit from the built-in on-chain capabilities provided by these frameworks to provide a more seamless experience when deploying on-chain agents as they can focus more on a well-designed front-end user experience. Furthermore, by leveraging blockchain and smart contracts as underlying infrastructure, on-chain agents can benefit from encryption tracks, such as having their wallet perform on-chain operations on behalf of users and ensuring incentive consistency.

Web3 Multi-Agility Framework Performance Indicators

Simplify Web3Workflows in

Although the Web2 proxy framework is becoming increasingly mature and popular, the proxy concept didn't gain attention in Web3 until the fourth quarter of 2024. Major players such as ElizaOS, Virtuals Protocol and RIG (each owns their own tokens) achieved substantial market cap, highlighting the strong demand for AI agents in Web3, rather than just speculative trading. The excitement reflected by the market caps of these tokens is not unfounded, as Web3 is still struggling to achieve mainstream adoption. Letting agents on blockchain perform on-chain operations independently has great potential to change the user experience. In addition to the efficiency that can be achieved, the problem of proxying in Web3 can be traced back to similar arguments about AI in blockchain, namely transparency and traceability, as well as advanced security features. Agent transactions are recorded on the blockchain, and users can easily track and verify actions taken by the agent. Below, we highlight some of the key areas that are best suited for proxy adoption.

DeFAI

On-chain transactions are inherently complex, requiring users to have at least a basic understanding of blockchain and Web3 wallets. This creates a bad user experience and remains a significant obstacle for non-encrypted native users. Although social logins have been widely adopted by various Web3 wallet providers recently, the development of account and chain abstractions remains slow and limited. When browsing the DeFi landscape, users still need to understand the concepts of gas fees, wallet addresses, and bridges. In contrast, OpenAI's recently launched Operator Agent requires users to perform simple natural language processing to execute transactions, abstracting multiple steps users must take through back-end proxy processing. Web3 shouldn't be any different, either, we believe that integrating AI proxy with various DeFi (DeFAI) protocols can facilitate easier user introductory and seamless experiences.

Virtuals Protocol recently launched the Agent Business Agreement, which sets a standardized approach to how agents communicate and interact with each other. This approach introduces a process involving 4 phases of request, negotiation, transaction and evaluation. The introduction of evaluators, smart contract-based custody and cryptographic verification is the core features of the framework to ensure that the delivered transactions meet the requirements of the task. Once all requirements are met, the smart contract trigger unlocks funds and delivers services, ensuring transactions can be conducted transparently and without trust. Agent Business Protocol is just an example of how a multiagent orchestration framework can help drive proxy interactions on-chain in a trustless and secure manner.

Olas Protocol demonstrates the practical application of DeFAI: its Pearl app store includes Mobius and Optimus agents, which use the Olas stack to automate DeFi policies on platforms such as Uniswap, Balancer, and Sturdy, covering networks such as Optimism, Base, and Mode. Olas Protocol's Mech market also acts as a proxy tool and plug-in switching center, allowing deployed proxy to outsource tasks through inter-agent communication. Another notable example is Questflow, which also proposes an intent-matching multiagent orchestration framework where users' requests are processed by orchestrators that identify the relevant agents and delegate the agent to these tasks through a task manager that supervises the execution of the agent's workflow. Agent creators can also receive fair compensation because agents are dispatched in the Deagent proxy registry.

Data ownership

In the huge proxy landscape and the large amount of on-chain data generated, on-chain analysis is becoming an increasingly valuable area, and many projects are seeking to provide data marking services (such as Sahara AI), tracking (Arkham Intelligence, Kaito), proof registry (EAS, BAS, etc.). As a user's right-hand assistant, the agent can contribute to the growing data landscape in Web3 by obtaining the user's permission, allowing users to receive fair rewards for their data contributions.

Game

In the Web3 gaming community, interest and demand for AI-enabled agents are growing. Game Agents can power non-player characters (NPCs) or manage in-game economy. They help create dynamic, responsive environments by performing tasks autonomously and responding to player actions. Notable projects in this area include Parallel’s WayFinder platform, which is building a knowledge graph that can be used by AI agents in different agent workflows in the game. Treasure DAO is another noteworthy example, which recently announced the upcoming launch of the MAGE proxy launch platform powered by ElizaOS, further moving towards the proxy-driven Web3 gaming landscape. Virtuals Protocol also launched Project WestWorld, an interactive simulation in Roblox that is powered by the G.A.M.E framework to autonomously interact and drive dynamic game narratives..

Other Use Cases

AI-driven DAO: Agents can distil lengthy proposals into easy-to-understand information that mainstream users can understand and vote, thereby enhancing the core spirit of decentralization.

Smart contract audit, network analysis, fraud detection: Agents can play a crucial role in debugging, often identifying potential risks faster than humans, thereby reducing security risks when combined with human intelligence.

Supply Chain Optimization: This can simplify and enable more cost-effective operations by using AI's predictive power and the transparency and security capabilities of blockchain.

Challenges and efforts for mature Web3 multiagent systems

Multiagent systems (MAS) in Web3 environments, where agents run on decentralized infrastructure and are often coordinated using smart contracts, face some limitations and challenges that may affect their design, deployment, and performance. Here are some of the hurdles that Web2 and Web3 agents may face:

Like a single LLM-based systems, multiagent systems are affected by the risk of model hallucination. When hallucinations are passed from one proxy to another, the risk of hallucination in multiagent systems can be more severe, exacerbating the problem. Poorly managed communication between agents will lead to suboptimal performance. Therefore, as we move towards a completely autonomous agent in the future, many frameworks still require some manual supervision.

Implement consensus and state synchronization between agents. In multiagent systems, in order to successfully complete tasks, agents must navigate in complex and layered multiagent systems to ensure consistency with overall tasks, their own responsibilities, and multiagent communication.

Proxys in Web3 also face scalability and latency issues as they run on the underlying blockchain and thus compete with other types of transactions. This could mean that we don’t see a full-on-chain orchestration of large proxy networks for the foreseeable future until the blockchain scalability challenge is resolved. Security and privacy challenges on blockchain are also unique in the Web3 environment, which adds complexity. However, this is slowly being resolved with emerging solutions such as Turnkey, which offers a TEE solution (AWS NitroEnclaves), where the agent can perform operations securely and verifiably. Phala Network also announced a partnership with GoPlus to enhance ElizaOS proxy using Phala's TEE capabilities and GoPlus security capabilities.

Multi-intelligent memory management. In a multiagent system, different agents perform different tasks and store different information. Therefore, to ensure successful delivery of the overall goal, it is helpful to reach an information consensus, while implementing a strong access control mechanism is critical, as some agents may be processing highly sensitive information. Failure to implement powerful security measures can lead to data privacy breaches and task execution failures.

The lack of comprehensive benchmarks and evaluation standards in certain fields such as science laboratory experiments, economics modeling and on-chain skills may hinder the rapid growth of the field.

Conclusion

The future of multi-agent frameworks is full of hope but also full of challenges, which highlights the long road ahead. Compared with the established and institutionally recognized Web2 multi-agent framework, the Web3 multi-agent framework is still in a relatively infancy with narrow production-ready use cases. Nevertheless, regulatory shifts and ongoing efforts to alleviate the above challenges are key catalysts for further adoption.

In addition, the growth of agent development tools (such as SendAI suite, Coinbase agent suite, ShellAgent codeless platform, Olas stack, etc.) is to simplify agent creation and scale agent use cases, continue to make progress, driving developer growth and new innovations. Advances in Web3 libraries such as GOAT SDK help extend the possibilities of operations implemented by proxy. Ultimately, with the development of technology and the maturity of these systems, we can expect agent workflows to become commonplace in on-chain interactions. Just like there are many Web2 multiagent frameworks, we want to see more proxy frameworks in Web3 that provide common and niche approaches.