Author: Kevin, the Researcher at BlockBooster
The AI Agent framework, as a key piece of the industry's development, may have the dual potential to drive technology implementation and ecosystem maturity. The popular frameworks in the market include: Eliza, Rig, Swarms, ZerePy, etc. These frameworks attract developers and build reputation through Github Repos. Similar to how "tokens" are issued as "libraries", these frameworks possess both the characteristics of waves and particles, just as Agent frameworks have both serious externalities and Memecoin features. This article will focus on interpreting the "wave-particle duality" of the frameworks and why Agent frameworks can become the final piece.
The externalities brought by the Agent framework can leave green shoots after the bubble bursts
Since the birth of GOAT, the impact of the Agent narrative on the market has been constantly increasing, like a kung fu master, with the left fist "Memecoin" and the right palm "industry hope", you will be defeated in one of the moves. In fact, the application scenarios of AI Agents are not strictly distinguished, and the boundaries between platforms, frameworks and specific applications are blurred, but they can still be roughly classified according to the preferences of tokens or protocols. However, according to the development preferences of tokens or protocols, they can be divided into the following categories:
Launchpad: Asset issuance platform. Virtuals Protocol and clanker on Base chain, Dasha on Solana chain.
AI Agent applications: Floating between Agent and Memecoin, with outstanding performance in memory configuration, such as GOAT, aixbt, etc. These applications are generally one-way output, with very limited input conditions.
AI Agent engines: griffain on Solana chain and Spectre AI on Base chain. griffain can evolve from read-write mode to read-write-action mode; Spectre AI is a RAG engine, on-chain search.
AI Agent frameworks: For framework platforms, the Agent itself is an asset, so the Agent framework is the asset issuance platform for the Agent, the Launchpad for the Agent. The representative projects currently are ai16, Zerebro, ARC and the recently discussed Swarms.
Other small directions: Comprehensive Agent Simmi; AgentFi protocol Mode; Falsification Agent Seraph; Real-time API Agent Creator.Bid.
Further discussing the Agent framework, we can see that it has sufficient externalities. Unlike the developers of major public chains and protocols who can only choose from different development language environments, the overall number of developers in the industry has not shown a growth rate corresponding to the increase in market capitalization. Github Repo is where Web2 and Web3 developers build consensus, and establishing a developer community here is more attractive and influential to Web2 developers than any "plug-and-play" package developed by a single protocol.
The 4 frameworks mentioned in this article have all been open-sourced: Eliza framework of ai16z has obtained 6200 stars; ZerePy framework of Zerebro has obtained 191 stars; RIG framework of ARC has obtained 1700 stars; Swarms framework of Swarms has obtained 2100 stars. Currently, the Eliza framework is widely used in various Agent applications and is the framework with the widest coverage. The development degree of ZerePy is not high, and the development direction is mainly on X, and it does not yet support local LLM and integrated memory. RIG has the highest relative development difficulty, but it can give developers the greatest freedom to achieve performance optimization. In addition to the team's launch of mcs, Swarms has not yet had other use cases, but Swarms can integrate different frameworks and has a large imagination space.
Furthermore, in the above classification, the separation of Agent engines and frameworks may cause confusion. But I think there is a difference between the two. First, why is it called an engine? Associating with search engines in real life is relatively fitting. Unlike the homogenized Agent applications, the performance of the Agent engine is above them, but at the same time it is completely encapsulated and adjusted through the API interface as a black box. Users can experience the performance of the Agent engine in the form of forks, but they cannot grasp the overall picture and customization freedom like the basic framework. Each user's engine is like generating a mirror image on a well-trained Agent, and it is to interact with the mirror image. The framework, on the other hand, is essentially to adapt to the chain, because when doing Agent frameworks, the ultimate goal is to integrate with the corresponding chain, how to define data interaction methods, how to define data verification methods, how to define block size, how to balance consensus and performance, these are the things the framework needs to consider. As for the engine, it only needs to fully fine-tune the model in a certain direction and set the relationship between data interaction and memory, and performance is the only evaluation standard, while the framework is not the case.
Evaluating Agent frameworks from the perspective of "wave-particle duality" may be a prerequisite for ensuring that we are on the right track
In the life cycle of an Agent's input and output, three parts are needed. First, the underlying model determines the depth and mode of thinking, then the memory is the customizable part, and after the basic model has output, it is modified according to the memory, and finally the output operation is completed on different clients.
Source: @SuhailKakar
To prove that the Agent framework has "wave-particle duality", the "wave" has the characteristics of "Memecoin", representing community culture and developer activity, emphasizing the attractiveness and propagation ability of the Agent; the "particle" represents the characteristics of "industry expectation", representing the underlying performance, actual use cases and technical depth. I will explain this from two aspects by taking the development tutorials of three frameworks as examples:
The quick assembly-style Eliza framework
1. Set up the environment
Source: @SuhailKakar
2. Install Eliza
Source: @SuhailKakar
3. Configuration file
Source: @SuhailKakar
4. Set Agent personality
Source: @SuhailKakar
Eliza's framework is relatively easy to get started with. It is based on TypeScript, which is a language familiar to most Web and Web3 developers. The framework is concise and not overly abstract, allowing developers to easily add the features they want. Through step 3, we can see that Eliza can be integrated with multiple clients, which can be understood as an assembler for multi-client integration. Eliza supports platforms like DC, TG and X, and also supports various large language models, allowing input through the above social media, LLM models for output, and built-in memory management support, so that any familiar developer can quickly deploy AI Agents.
Due to the simplicity of the framework and the richness of the interfaces, Eliza greatly reduces the access threshold and realizes a relatively unified interface standard.
The one-click usage-style ZerePy framework
1. Fork the ZerePy repository
Source: https://replit.com/@blormdev/ZerePy?v=1
2. Configure X and GPT
Source: https://replit.com/@blormdev/ZerePy?v=1
3. Set Agent personality
Source: https://replit.com/@blormdev/ZerePy?v=1
The performance optimization-style Rig framework
Taking the construction of a RAG (Retrieval-Augmented Generation) Agent as an example:
1. Configure the environment and OpenAI key
Source: https://dev.to/0thtachi/build-a-rag-system-with-rig-in-under-100-lines-of-code-4422
2. Set up the OpenAI client and use Chunking to process PDFs
Source: https://dev.to/0thtachi/build-a-rag-system-with-rig-in-under-100-lines-of-code-4422
3. Set up the document structure and embedding
Source: https://dev.to/0thtachi/build-a-rag-system-with-rig-in-under-100-lines-of-code-4422
4. Create vector store and RAG agent
Source: https://dev.to/0thtachi/build-a-rag-system-with-rig-in-under-100-lines-of-code-4422
Rig (ARC) is an AI system building framework based on the Rust language, oriented towards LLM workflow engines. It aims to solve more fundamental performance optimization problems, in other words, ARC is an "AI engine toolbox" that provides AI invocation, performance optimization, data storage, exception handling and other backend support services.
Rig's focus is on the "invocation" problem, helping developers better choose LLMs, better optimize prompts, more effectively manage tokens, and how to handle concurrency processing, manage resources, and reduce latency, with the emphasis on how to "use it well" in the collaboration process between AI LLM models and AI Agent systems.
Rig is an open-source Rust library aimed at simplifying the development of LLM-driven applications (including RAG Agent). Since Rig is more open, it requires higher demands on developers and a deeper understanding of Rust and Agents. This tutorial is the most basic configuration process for the RAG Agent, where RAG enhances LLM by combining it with external knowledge retrieval.
Unified LLM interface: Supports consistent APIs for different LLM providers, simplifying integration.
Abstract workflow: Pre-built modular components allow Rig to undertake the design of complex AI systems.
Integrated vector store: Built-in support for vector storage, providing high-performance in search-based Agents like RAG Agent.
Flexible embedding: Provides easy-to-use APIs for handling embeddings, reducing the difficulty of semantic understanding in the development of search-based Agents like RAG Agent.
Compared to Eliza, Rig provides developers with additional space for performance optimization, helping them better debug the invocation and collaboration optimization of LLM and Agents. Rig is driven by Rust's performance, leveraging Rust's advantages of zero-cost abstraction and memory safety, high-performance, low-latency LLM operations, and can provide more freedom at the underlying level.
Decomposing the Swarms framework
Swarms aims to provide an enterprise-grade production-ready multi-Agent orchestration framework, with the website providing dozens of workflow and Agent parallel-serial architectures, here introducing a small part of them.
Sequential Workflow
Source: https://docs.swarms.world
The sequential Swarm architecture processes tasks in a linear order. Each Agent completes its task before passing the result to the next Agent in the chain. This architecture ensures ordered processing and is very useful when tasks have dependencies.
Use cases:
Each step in a workflow depends on the previous step, such as an assembly line or sequential data processing.
Scenarios that require strict adherence to the order of operations.
Hierarchical architecture:
Source: https://docs.swarms.world
Implements top-down control, with the parent Agent coordinating tasks between the child Agents. Agents execute tasks simultaneously and then feed their results back into the loop for final aggregation. This is very useful for highly parallelizable tasks.
Spreadsheet-like architecture:
Source: https://docs.swarms.world
For managing large-scale swarms of multiple agents working simultaneously. It can manage thousands of agents, each running on its own thread. It is an ideal choice for supervising the output of large-scale agents.
Swarms is not only an Agent framework, but can also be compatible with the aforementioned Eliza, ZerePy and Rig frameworks, using a modular approach to maximize the release of Agent performance in different workflows and architectures to solve corresponding problems. Swarms' conception and developer community progress are not problematic.
Eliza: Strongest in usability, suitable for beginners and rapid prototyping, especially suitable for AI interactions on social media platforms. The framework is concise, easy to integrate and modify, suitable for scenarios that do not require excessive performance optimization.
ZerePy: One-click deployment, suitable for rapid development of Web3 and social platform AI Agent applications. Suitable for lightweight AI applications, the framework is simple, configuration is flexible, and suitable for rapid deployment and iteration.
Rig: Focuses on performance optimization, especially excelling in high concurrency and high-performance tasks, suitable for developers who need fine-grained control and optimization. The framework is more complex and requires some Rust knowledge, suitable for more experienced developers.
Swarms: Suitable for enterprise-level applications, supporting multi-Agent collaboration and complex task management. The framework is flexible, supports large-scale parallel processing, and provides various architectural configurations, but its complexity may require a stronger technical background to use effectively.
Overall, Eliza and ZerePy have advantages in usability and rapid development, while Rig and Swarms are more suitable for professional developers or enterprise applications that require high performance and large-scale processing.
This is the reason why Agent frameworks have the "industry hope" feature. The aforementioned frameworks are still in the early stages, and the urgent task is to seize the first-mover advantage and establish an active developer community. The performance of the framework itself and whether it is lagging behind the popular Web2 applications are not the main contradictions. Only frameworks that constantly attract developers can ultimately prevail, because the Web3 industry always needs to attract the attention of the market. Even if the framework's performance is strong and the fundamentals are solid, if it is difficult to get started and no one cares, it will be putting the cart before the horse. On the premise that the framework itself can attract developers, the framework with a more mature and complete token economic model will stand out.
The fact that Agent frameworks have the "Memecoin" feature is also very understandable. The aforementioned framework tokens do not have a reasonable token economic design, the tokens have no use cases or the use cases are very single, there is no verified business model, and there is no effective token flywheel. The framework is just a framework, and there is no organic integration with the tokens, and the growth of token prices, apart from FOMO, is difficult to obtain support from the fundamentals. There is not enough moat to ensure stable and lasting value growth. At the same time, the aforementioned frameworks themselves also appear to be somewhat rough, and their actual value and current market value do not match, so they have strong "Memecoin" characteristics.
It is worth noting that the "wave-particle duality" of Agent frameworks is not a shortcoming, and it cannot be roughly understood as neither a pure Memecoin nor a token with use cases. As I mentioned in the previous article, the lightweight Agent covers the ambiguous Memecoin facade, and community culture and fundamentals will no longer become contradictions, a new asset development path is gradually emerging. Although Agent frameworks initially have bubbles and uncertainties, their potential to attract developers and drive application landing cannot be ignored. In the future, frameworks with a complete token economic model and a strong developer ecosystem may become the key pillars of this track.
About BlockBooster
BlockBooster is an Asia-based Web3 venture studio supported by OKX Ventures and other top-tier institutions, committed to becoming a trusted partner for exceptional entrepreneurs. Through strategic investment and in-depth incubation, we connect Web3 projects with the real world and help high-quality startups grow.
Disclaimer
This article/blog is for reference only, representing the author's personal views and not the position of BlockBooster. This article does not intend to provide: (i) investment advice or investment recommendations; (ii) an offer or solicitation to buy, sell or hold digital assets; or (iii) financial, accounting, legal or tax advice. Holding digital assets, including stablecoins and Non-Fungible Tokens (NFTs), is extremely risky, with high price volatility and the possibility of becoming worthless. You should carefully consider whether trading or holding digital assets is suitable for you based on your financial situation. For specific issues, please consult your legal, tax or investment advisor. The information provided in this article (including market data and statistics, if any) is for general reference only. Reasonable care has been taken in compiling these data and charts, but no responsibility is accepted for any factual errors or omissions contained herein.
