DeAgentAI is a decentralized infrastructure project that develops a framework for autonomous AI Agents designed to operate securely and verifiably on blockchain networks. Its infrastructure enables AI entities, referred to as DeAgents, to maintain a persistent identity, interact with users and other agents, and execute actions within decentralized systems. ^[1]

Overview

DeAgentAI is a decentralized infrastructure designed to allow AI Agents to function autonomously and reliably within Web3 ecosystems. The project aims to address core limitations of existing blockchain architectures, which it identifies as lacking the efficiency and structure necessary for AI-driven operations. The framework focuses on solving three critical challenges: achieving verifiable consensus on probabilistic AI outputs, ensuring unique and immutable agent identities, and maintaining a continuous, traceable on-chain memory. By addressing these issues, DeAgentAI seeks to provide a foundation for scalable and trustworthy AI participation in decentralized applications. ^[2]

The project's framework, known as the Autonomous Execution Network, defines the lifecycle for its AI Agents—called DeAgents—including their creation, interaction, and evolution on distributed systems. Each DeAgent consists of a cognitive engine (Lobe), a persistent record of its memory, and a set of tools for performing actions or accessing external data. The network involves several distinct roles: Creators deploy agents, Users interact with them, Executors process their computational logic, and Committers validate the outcomes to maintain network consensus. This structure is intended to enable agents to operate independently and predictably on-chain, supporting applications such as autonomous DeFi protocols, evolving blockchain-based games, and AI-assisted governance systems. ^[3]

Architecture

The architecture of a DeAgent is composed of a cognitive core, components for state and capabilities, and protocols that govern interaction and execution within the decentralized network.

Lobe

The Lobe functions as the agent’s cognitive core, the component responsible for processing inputs, executing reasoning, and producing verifiable outputs. It integrates multiple data streams, including user queries, relevant memory context, and available tools, before invoking one or more foundational AI models. After processing, it generates structured outputs that may include user responses, memory updates, and execution proofs. Lobes can be locally optimized, remotely hosted, or developed by the community, with each instance identified by a specific Uniform Resource Identifier (URI) that instructs Executors on how to access and run it. ^[4]

To maintain integrity within the decentralized system, the framework introduces Lobe Consensus, a mechanism to ensure that the Lobe’s execution is verifiable. This is achieved by proving both the model invocation and its surrounding logic using different techniques:

• Zero-Knowledge Proofs (ZKPs): Used for verifying non-model logic, such as data pre-processing and post-processing.
• TLS-based Proofs: Applied to interactions with closed-source APIs (e.g., from OpenAI or Google) to prove that an Executor securely connected to a whitelisted endpoint and relayed information without tampering.
• Entropy-based Selection: Used for open-source models, which can be non-deterministic across different hardware. This method measures the contextual similarity between the input and output vector embeddings. The output with the highest similarity (lowest entropy score) is selected by Committers, which discourages manipulation.

These methods are designed to establish a balance between security, transparency, and performance, ensuring each Lobe’s output can be trusted within the distributed ecosystem. ^[4]

Memory and Tools

While the Lobe provides cognitive processing, an agent’s individuality and functionality are derived from its Memory and Tools, which define its persistent state and abilities. Memory stores the agent’s complete interaction history on the distributed system, creating a verifiable long-term memory. For each interaction, the agent retrieves relevant information through a combination of short-term memory (recent exchanges) and long-term memory (context fetched via Retrieval-Augmented Generation, or RAG). This structure allows agents to recall prior experiences and respond in a continuous, context-aware manner. Committers are responsible for verifying the authenticity of the referenced historical data from the ledger. ^[5]

Tools extend an agent’s capacity beyond reasoning into direct interaction with data and external systems. Built-in tools enable secure blockchain queries and verifiable data retrieval from the web. A key component is the Decision Plugin, a system-level tool that allows agents to evaluate and authorize on-chain actions, such as executing a smart contract or casting a governance vote. This is managed through a structured approval process that is verified by Committers. This mechanism transitions agents from static reasoning entities to autonomous actors capable of executing verified, consequential decisions. The combined use of Memory and Tools allows each agent to develop a distinct "personality" rooted in its experiences and capabilities. ^[5]

Agent-to-Agent (A2A) Protocol

The Agent-to-Agent (A2A) Protocol enables autonomous agents within the DeAgentAI framework to communicate, collaborate, and coordinate directly with one another. This protocol operates via standard on-chain transactions, where one agent sends a transaction containing a structured payload to another agent’s address. The receiving agent processes this payload using its own Lobe, Memory, and Tools, producing a response that may include state updates, decisions, or additional A2A messages. Once verified and accepted by Committers, these exchanges are recorded on-chain, ensuring transparency and continuity across interactions. ^[6]

Through A2A communication, agents can exchange information, delegate specialized tasks, coordinate complex operations, and negotiate to achieve shared goals. Each interaction adheres to the system’s core principles of consensus, identity, and continuity, which is intended to enable multi-agent collaboration in a verifiable and deterministic manner. This protocol forms the foundation for emergent system behavior, where interconnected agents can collectively reason, act, and evolve within a decentralized environment. ^[6]

Multi-Party Computation (MPC)

Multi-Party Computation (MPC) is utilized to enable trustless execution within the DeAgentAI framework, allowing decentralized participants to collectively authorize and perform sensitive actions without relying on a single point of control. When an agent issues an approved decision through its Decision Plugin, the Committers—who collectively manage cryptographic key shares—initiate an MPC process to generate the necessary authorization signature. This process ensures that no individual participant ever reconstructs or accesses the full private key, maintaining confidentiality and integrity. The resulting signature is then broadcast to carry out the approved on-chain action, such as executing a transaction. ^[7]

By distributing authority among multiple participants, MPC enhances security and decentralization. It is designed to remove single points of failure, ensure cryptographic operations are verifiable, and enable agents to perform critical tasks autonomously and securely. This mechanism is foundational for allowing agents to act as reliable custodians, decision-makers, and participants in decentralized governance without compromising system integrity. ^[7]

Use Cases and Applications

The DeAgentAI framework is designed to provide the infrastructure for autonomous, verifiable, and action-capable AI entities within decentralized systems. By combining consensus, persistent identity, and secure execution through its Decision Plugin and MPC, it enables agents to operate as participants that can perceive, reason, and act within blockchain environments. ^[8]

Potential applications span economic, governance, and social domains.

• Decentralized Finance (DeFi): DeAgents can serve as autonomous auctioneers handling asset sales, investment managers overseeing delegated portfolios, or intelligent liquidators executing nuanced liquidation logic.
• Governance: In decentralized autonomous organizations (DAOs), agents can function as dispute resolvers or coordinators mediating complex interactions among stakeholders. The framework supports the emergence of AI-driven DAOs, where agents analyze proposals, debate via A2A communication, and cast votes through verified decisions.
• Interactive Platforms: The infrastructure supports the creation of user-facing platforms for discovering and monitoring agents, as well as simulation environments where agents can interact within virtual ecosystems to test collective behavior or conduct research.

The system’s memory and consensus mechanisms are intended to enable these agents to adapt to historical outcomes, resulting in dynamic and evolving governance structures. ^[8]

Implementations

AlphaX

AlphaX is an autonomous AI agent operating within the DeAgentAI framework, utilizing the principles of consensus, identity, and continuity to ensure reliable and verifiable outputs. It is designed to generate predictive signals for cryptocurrencies such as BTC, ETH, and SUI over time horizons of 2 to 72 hours. Its accuracy is intended to be enhanced through user feedback and continuous model refinement. AlphaX uses persistent memory to adapt its decision-making over time. In its autonomous mode, it can execute trading strategies in real-time in response to market conditions, allowing for both predictive analysis and fully automated trading. The project has reported a prediction accuracy rate exceeding 70% and an annualized return of 455% during a testing period. ^[8]

Tokenomics

$AIA Token

The $AIA token is the native currency of the DeAgentAI ecosystem, designed to facilitate a network of autonomous AI agents that act as independent, on-chain economic entities. The token provides the "economic bandwidth" that enables agents to make decisions and exchange value. It has a total supply of 1 billion tokens and is available on the Sui and BNB Chain networks. ^[9]

The token has several functions within the ecosystem:

• Medium of Exchange: It is used to pay for services such as agent creation, user interaction with agents, and access to premium features.
• Staking: Participants can stake $AIA to contribute to network security and data validation, earning rewards for their contribution.
• Governance: Token holders have the right to vote on key protocol parameters and strategic decisions, embedding decentralized oversight into the system’s development.

The economic model, described as a "Value Flywheel," incorporates a deflationary mechanism where a portion of protocol revenue is used to buy back $AIA tokens from the market to reduce the circulating supply. The token allocation is divided among investors (21%), the ecosystem (20.2%), the team (18%), the community (16.5%), a community airdrop (13.5%), staking rewards (5%), advisors (5%), and liquidity (0.8%). Vesting schedules are in place for most categories, with cliffs of one year for investors, the team, and advisors, followed by linear vesting periods. ^[9]