Bounce Research: What is Agent Virtual Machine?

Bounce Brand
4 min readMar 27, 2024


Have you ever pondered how AI agents navigate the complexities of decentralized systems while ensuring efficiency and autonomy? In this article, we explore the concept of Agent Virtual Machine (AVM). Recollecting our previous research of an innovative solution called Decentralized Nodes for AI Agents (DNA), we shed light on the crucial role of decentralized nodes in autonomous AI operations within Web3 environments. Furthermore, our exploration of the infrastructure behind DNA technology underscored the significance of runtime providers ensuring the efficiency of AI agents’ performance. AVM serves as the foundation for constructing DNA, providing an efficient runtime environment for AI agents. Let’s unravel the inner workings of AVM and its pivotal role in shaping the future of decentralized AI.

What is Agent Virtual Machine?

An Agent Virtual Machine is a virtual machine designed to enable AI agents to perform tasks within the virtual infrastructure they serve.

AVM bears similarities to Ethereum Virtual Machine (EVM), both pivotal in their respective domains. EVM lays the foundational environment for executing smart contracts, ensuring their operational consistency across blockchain nodes. Likewise, AVM caters to AI Agents by offering a dedicated runtime environment, equipped with essential resources required for their enduring functionality.

AVM’s capabilities make it a great fit for bridging artificial intelligence and blockchain technology, illustrating the adaptability and potential of virtual machines in supporting sophisticated computational frameworks and the continuous evolution of Agents within their lifecycle.

How does AVM work?

AVM is designed based on four pivotal elements: State Management, Action, Reaction, and Multimodal Input/Output (IO). These components collectively empower AI agents to persistently manage data states, perform autonomous tasks, respond dynamically to stimuli, and process diverse inputs and outputs.

  1. State Management:

In the context of AVM, “State” refers to information that can be persistently stored, dynamically modified, and influences behavior throughout an AI Agent’s operational process. This includes knowledge bases, memory, and concurrency management, all managed via the “State” module.

AVM provides an environment tailored for state storage, incorporating a structured database for long-term state storage, runtime memory for short-term state storage, and a vector database for storing data related to knowledge bases. Developers are thereby enabled to efficiently perform read and write operations for memory, knowledge bases, and concurrent states, leveraging AVM for advanced and professional AI language processing.

2. Action:

In the context of AVM, “Action” refers to a series of actions that an AI Agent autonomously initiates under the control of objectives or other conditions set by developers, such as browsing news, updating databases, sending messages to Slack, etc. AVM enables Agents to perform various conditional proactive actions throughout their lifecycle. Developers can effortlessly control the triggering conditions and intervals, reiterate and terminate the logic of AI Agent’s proactive actions.

3. Reaction:

In the context of AVM, “Reaction” refers to an AI agent’s response to external information. These reactions are triggered either by changes in monitored external information or by user requests sent to the Agent, such as chat requests. The widely utilized chat functionality of ChatGPT is an example of such a reactive action. The Agent Virtual Machine (AVM) enables Agents to implement the most common responses and provides a more advanced abstraction called the “Agent Hook.” This hook triggers the Agent’s reactive actions when specific parameters change, facilitating the development of high-performance trading bots or news bots.

4. Multimodal I/O(Input/Output):

As of 2024, the vast majority of large AI models lack the capability for multimodal input or output. For developers, implementing multimodal conversions is a cumbersome task, involving format decoding, data compression, and the synchronization of temporal sequences, among other challenges.

AVM supports the management of multimodal I/O, thereby controlling the permissible modalities for input and output. Furthermore, AVM facilitates the conversion between information modalities. Developers can use the interfaces provided by AVM to transform all image input information into text, enabling multimodal input and output on the foundation of a purely linguistic model.

How does AVM empower Decentralized Nodes for AI Agents?

The innovation solution we previously introduced, known as Decentralized Nodes for AI Agents, is developed based on the foundational technology of AVM to bridge AI agents and Web3 applications. Powered by AVM’s runtime provider functionalities, DNA allows AI agents to operate persistenty and securely, perform autonomous decision-making, thereby allowing these agents to carry out tasks proactively and with minimal human intervention.

Thanks to the the robust infrastructure provided by AVM, developers can easily create AI agents that are high-performance, secure, and versatile on Web3 platforms. This eliminates the necessity for manual management of intricate runtime resources.

In conclusion, our exploration of AVM has demonstrated its foundational role in the DNA solution that aims to advance decentralized AI operations within Web3 environments. From understanding its concept and key components to comparing it with blockchain systems like EVM, hope you have gained some insights on AVM and its relation to the DNA technology. Stay tuned for our next article as we delve into the challenges faced by traditional Layer1 solutions, DAOs, and cloud services in supporting AI Agent operations, showcasing why DNA stands out as an enhanced solution to bridge AI and Web3.

About Bounce Brand

Bounce Brand is a prominent decentralized platform dedicated to delivering: BounceBit, building a BTC restaking infrastructure; Bounce Auction, onchain decentralized auctions; BounceClub, an all-in-one Web3 space simplifying smart contract deployment on BounceBit — all tailored for the Bitcoin ecosystem. Committed to enhancing Bitcoin’s smart contract functionality, Bounce Brand strives to empower transformative growth within the blockchain landscape.


Website | App | Docs | Telegram Announcement | Telegram Community | X(Former Twitter) | YouTube | Discord | Zealy | LinkeIn



Bounce Brand

An experiment. Decentralized Auction. Swap or get bounced.