AI agents at the intersection of AI and crypto have become a top trend in 2025. Google introduced AI agents as software systems engineered to go beyond traditional AI models by interacting with external systems, making decisions, and completing tasks without human intervention.
While LLMs or AI models like GPT-4 or Google’s Gemini operate based on what they have learned from their training data, AI agents interact with systems, learn, and execute tasks. AI agents boast reasoning capabilities and are poised to change how firms operate, compete, and innovate.
Two frameworks, Eliza and GAME, have dominated conversations around AI agent frameworks stemming from the AI revolution, demand for autonomous AI agents, popularity among developers, and expansion of their ecosystems.
While Eliza and GAME are rather distinct AI agent frameworks with different applications, they both integrate and foster user interaction on Twitter.
Eliza is an AI agent framework developed by AI16Z (now ElizaOS) and launched in October 2024. It lets developers build and deploy unique AI agent characters that can interact with users on social media platforms.
Conversely, AI agents are initially created on the Virtuals Protocol but rely on the GAME framework for agentic capabilities—an AI agent’s ability to act autonomously. GAME provides an execution environment to test an agent’s capabilities.
This piece investigates the two leading AI agent frameworks using common competitive factors. Let’s begin.
Eliza is an open-source conversational and web3-friendly framework that facilitates the deployment of autonomous AI agents. It fills the gap and integrates web3 applications seamlessly into AI agents so they can perform tasks like reading, writing blockchain data, and interacting with smart contracts.
Built by the developers behind AI16Z, Eliza uses the popular TypeScript language and allows users to retain control. Eliza is ideal for agents that are embedded in the web. It is compatible across multiple platforms, powering interaction with different chains and social media platforms. Eliza also boasts superior and flexible plugins to extend AI capabilities and promote innovation.
Eliza also supports several networks like Ethereum, Solana, and TON. It integrates well with AI model providers such as Llama and OpenAI, as well as social platforms such as X and Discord.
Eliza aims to become a fully autonomous AI system that can plan and execute tasks without human interference. It integrates extensive plugins like RAG and text to image/videos/3D; hence, Eliza can support multimodal interactions like text, voice, and media.
Some of the key features of Eliza include intent recognition—understanding the objective of a user request. The primary mechanism comprises a hierarchical action structure that defines intent by a primary identifier plus a collection of semantic similies designed to recognize users’ intent across various linguistic expressions.
While the Virtuals protocol provides a custom process for creating AI agents (allowing users to specify the agent's name, image, token ticker, and character description), GAME uses an Agent Sandbox (a test environment) to define key characteristics.
GAME can be integrated into applications either as a preconfigured setup (plug and play) or as a game-as-a-service via API calls. Its services can be accessed via the GAME cloud (a hosted service) or the GAME SDK; both require an API key.
GAME enables the building and deploying of autonomous AI agents to gaming, social media, and entertainment platforms. It's a decision-making engine for powering agents in diverse environments. When an agent is provided with information, GAME enables it to plan actions and take decisions.
The GAME (Generative Autonomous Multimodal Entities) framework affords a structured approach to the management of AI agents' behavior, letting users create a custom API for their agents using the GAME SDK. Users can configure an agent’s capabilities, run simulations, and test agents’ behavior before deployment.
Using the GAME SDK, users can achieve more control and integration with third-party apps. GAME SDK allows great customization so users retain full control over their AI agents. They can configure their environments and define how agents interact with each other and how they respond to different situations.
In short, the GAME framework by Virtuals offers a low barrier of entry and ease of use, allowing anyone to ramp up an AI agent with agentic capabilities. Powered by the GAME SDK, users can customize their environment and define how components interact with each other and respond to various situations.
This section compares both frameworks by architecture, use cases, ease of use, customization of models, tech stack, token value accrual, adoption, and limitations.
Eliza
Eliza’s modular design structure allows for the free addition of plugins and customization of AI agents without altering the core runtime. Its flexible structure comprises a core runtime (where AI agents run) and four components, i.e., Adapter, Character, Client, and Plugin. The Adapter is responsible for data integration, the Character defines the agent’s personality, the Client manages message interaction, and the Plugin affords universal functions.
Overall, Eliza’s architecture is composed of several components:
Eliza's plugin architecture is flexible for easy extension of AI agents' capabilities. It supports the collaborative development of modular and open-source software and extends Eliza AI agents' capabilities in various applications, including content creation, blockchain support, essential services, and DeFi. The Solana plugin is an example of Eliza’s plugin system suitable for developers trying to integrate Solana blockchain functionalities.
GAME
GAME’s architecture consists of a high-Level Planner (HLP) and a Low-Level planner (LLP). HLP is driven by the agent definition (goals and character cards) and information (world info, agent state, and location); they essentially control agent thinking and decision-making. The LLP comprises the functions that determine the action/skill provided for the agent’s execution.
Eliza:
Eliza plugs into in-demand applications such as DeFi, token management, and swapping; an example is the Solana Plugin built by Eliza for interacting with the Solana Blockchain. Eliza integrates well with multiple social media platforms, i.e., X, Telegram, and Discord. Eliza can power social interaction. It navigates the social media space and reduces the information overload problem. It can sift and dissect information from KOLs to produce insights. Another use case is image generation with the framework supporting image generation configuration and response handling.
GAME:
GAME is applicable for gaming environments, X(Twitter), and multi-agent interactions. They may come with custom functions/configurations; however, developers can adjust or add further customization to their agents. Typical functions may include tweeting, liking, and searching X profiles for tweets. Developers can use Agent Sandbox to simulate various elements before actual deployment. GAME supports multi-agent systems like SANTA, VADERAI. Applications can make API calls to access the decision-making capabilities of GAME, i.e., Game-as-a-Service.
Eliza:
Eliza’s design principles are built around the demands of Web3, such as easily pluggable components and extensions, familiar tools and frameworks, and a less complex internal implementation. Eliza integrates well with AI models and add-ons to widen access to advanced AI functionalities such as NLP, multi-agent simulation, and memory management.
While Eliza tries to reduce the entry barrier for developers to build complex AI applications, it is more suited for technical users. GAME: GAME is designed to be a no-code toolkit that makes for high ease of use, whether for building autonomous AI agents or an app that requires AI for decision-making. Users only need an API key to get started. It is suited for almost anyone or non-technical users.
Eliza:
Eliza offers full customization and control. Users are required to define the agent’s character when getting started; this requires defining the agent’s knowledge, personality, and behavior. The Eliza character components are essentially JSON files containing configurations about the AI agent’s personality and behavior. To define a character, developers must specify behavior, capabilities, and interaction, among other details.
GAME:
The GAME framework offers low-code model customization with the agent description or character card. Developers can define the AI agent's personality, i.e., prompt design. They can specify the agent's descriptions (backstory and appearance), personality, relationship, preferences, tone, and style. Alternatively, developers can use the CO_STAR framework (i.e., context, objective, style, tone, audience, and response) to customize their agents.
Eliza:
Developers must install Python, Node.js, and PnPm to get started with Eliza. They would also require an OpenAI API key and a current web browser.
Eliza’s SDK is available in TypeScript.
Every Eliza agent runs a runtime that affords an environment for agent-specific code to execute. The IAgentRuntime interface specifies the properties, methods, and events for managing the events runtime, or the structure that the class must follow. While the Agent Runtime class handles the implementation of the interface and manages the agents' core functions.
GAME:
GAME’s SDK is available in Python and TypeScript.
Eliza:
While Eliza does not have its own token, value accrual will occur via the tokens of other platforms where the Eliza framework is integrated.
GAME:
Agents on the Virtuals protocols can interact (API call) or query the GAME engine for decision-making. Whenever GAME processes a request, it uses resources, which can attract charges.
GAME isn’t limited to the virtual ecosystem. But as the ecosystem grows and more developers leverage the agentic capabilities of the GAME framework, it is expected that more value will naturally flow to the GAME token. On the Virtuals protocol, developers are charged an agent creation fee and an inference fee for AI agent actions.
Eliza:
In January, Eliza (AI16Z's framework) became a number one trending repo globally; it gained over 10.2k+ stars on GitHub, 2.7k+ forks, and over 271 contributors. Furthermore, the project attracted a research partnership agreement with Stanford University's Future of Digital Currency Initiative (FDCI) to drive the adoption of Eliza’s technology.
GAME:
GAME has greatly benefited from the rise of the virtual ecosystem. By late December, Virtuals led with over 2000 AI agents launched on the platform and captured over $60 million in protocol revenue. In the same month, the Virtuals team reported over 200 projects built using GAME and 150 requests sent daily. Requests in this context can be API calls, web requests, or interactions.
Eliza:
The Eliza team did acknowledge the limitations of its runtime design, which may need some refining to balance the computational loads of multiple agents; this alludes to a scalability problem that could arise as Eliza handles more complex tasks. Other limitations include limited support for popular languages like Python and Rust.
GAME:
GAME framework’s specialization in the social media and gaming context might limit broader applications; hence, developers looking to build might face technical limitations.
Wrapping up
Eliza boasts superior tech than the GAME framework; its extensive plugin system, cross-platform compatibility, and greater model customization make it a top choice. Eliza offers a steeper learning curve but a broader use case across SMM, DeFi, and image generation. While Eliza’s adoption is solid, its tech superpower particularly stands out.
Conversely, the GAME framework boasts a superior growth rate, benefiting from the Virtuals’ market dominance as its agentic layer. GAME is well positioned to gain network effects as more projects launch on the virtual ecosystem. While specialized in gaming and social interactions, GAME offers great ease of use with a no-code toolkit allowing just anyone to launch their own agent.
It's a case of tech versus rapid growth rate, with Eliza leading the tech side while GAME gaining faster adoption thanks to low-code and the Virtuals protocol. As the AI agent space continues to evolve, it will be interesting to see what greater functionalities both platforms offer developers to ship the best products and the kind of traction they manage to sustain in the market.
Resources
https://docs.game.virtuals.io/
https://venturebeat.com/ai/google-maps-the-future-of-ai-agents-five-lessons-for-businesses/
https://x.com/miratisu_ps/status/1878359036279890081
https://x.com/miratisu_ps/status/1878359036279890081
https://x.com/GAME_Virtuals/status/1876641516304920876
https://learn.bybit.com/ai/what-is-game-by-virtuals/
https://x.com/DoctorDeFi/status/1879252789873774665
https://x.com/virtuals_io/status/1859200652993585607
https://learn.bybit.com/ai/what-is-game-by-virtuals/
https://x.com/virtuals_io/status/1872215978312135023
https://elizaos.ai/articles/eliza-foundation
https://x.com/joinzo/status/1878233275287921041
https://x.com/0xwitchy/status/1876619754997518773
https://elizaos.ai/ https://x.com/DeFlow_Intern/status/1879122843037999472
https://elizaos.github.io/eliza/docs/faq/
https://x.com/virtuals_io/status/1874111116403761316
https://arxiv.org/pdf/2501.06781
https://hackmd.io/@metamike/eliza-agent-weaver
https://www.coingecko.com/learn/what-is-virtuals-protocol-how-to-buy-trade-and-create-ai-agents