NEAR is the blockchain for AI. Over the past year and a half the focus and scope of that mission has expanded significantly: AI is now becoming autonomous, decentralized, private, secure, and most importantly user-owned. This update synthesizes the most important AI developments in the NEAR Ecosystem across the many different teams working to build out the blockchain for AI. As we enter the growth phase of the AI Revolution, NEAR is a reminder that AI should work to become accessible and monetizable for everyone - and that blockchain is the ultimate coordination and trust mechanism between different stakeholders with different interests. 

Here is the state of AI on NEAR in 2025: 

Foundations: Decentralized Confidential Machine Learning (DCML)

Earlier this year, in 2025, the NEAR AI team led by Illia Polosukhin and Alex Skidanov published a paper on Decentralized Confidential Machine Learning (DCML). While complex and technical, the thesis around DCML proposes a business model for user-owned AI leveraging the NEAR blockchain as a coordination mechanism between GPU compute providers, model creators, agent builders, evaluation systems, and users looking to own and monetize their own agents and AI assistants. DCML is ultimately the foundational framework for integrating the various components of the AI ecosystem on NEAR in a manner that promotes privacy, verifiability, and ownership of core AI primitives. With DCML, the blockchain for AI can deliver decentralization and accessibility first. 

The core products building out the vision of DCML are multiple and rapidly developing:  

• The NEAR Agent Hub: The NEAR Agent Hub is the foundation for evaluating, utilizing, and connecting different agents on NEAR. Designed to host any agent, regardless of the framework or programming language used to create them, Agent Hub offers a comprehensive suite for monitoring your agents, including in-depth analytics tailored specifically to AI agents. Most importantly, Agent Hub provides an LLM gateway (aka an "auto" model picker) that spans multiple providers, enabling users to achieve 99.9% uptime when deploying an agent.Officially in beta, the agent hub offers access to a variety of agents including a private chatbot. 

• Fine-Tuning and Model Training: DCML enables multiple parties to coordinate and get rewarded via NEAR smart contracts for fine-tuning models, creating and testing benchmarks, data sourcing, cleaning, and more. ‘User-Owned AI models’ refer to fully private, accessible, and custom trained LLMs that are run on NEAR cloud and monetize participation and access in $NEAR token. More specifically, a number of niche and industry specific small language models (SLMs) and large language models (LLMs) - from financial modeling AI to proprietary data sets, to geographic or subject-specific inputs - can be trained in a decentralized and private manner on NEAR, and then monetized for the benefit of the owners and creators. This framework for fine tuning and model training is positioned to host a number of training competitions in 2025 and 2026, and begin the training of industry-specific AI models.

• Evaluation and Benchmarking of Agents: Forest AI is the first (of a number of teams) in the NEAR Ecosystem currently focused on improving benchmarking and evaluations among agents and agentic frameworks. Their mission is to ensure that no cheating is done by agent builders, when establishing agent performance or evaluating agent readiness for commercial purposes. Their products aspire to solve the problem of cheating with AI, such that all agents in the NEAR ecosystem accurately reflect their genuine capabilities, such that they can be deployed and monetized fairly.

Quality Assurance: Data Storage, Labelling, Cleaning and Synthetic Data: 

Functional models require reliable data for strong performance. This data must be safely stored, correctly labelled, cleaned, and even artificially generated to improve model performance and agent capacities. The NEAR ecosystem has a number of teams focusing on multiple facets of the AI Data problem: 

• Private Data Storage: Private data storage with xTrace as well as other select partners, is all about managing Agent and user identity on-chain for ensuring safe memory and context when using an AI application. xTrace is developing a unified account model using NEAR Accounts so data can live under one programmable on-chain identity with flexible permissions for different applications to gain access. NEAR users and developers will have private vector stores available to personalize agents with their own personal data vault. Each NEAR Account can store access control logic and data references on-chain (pointers to your AI context/memory), grant granular permissions to different apps, and even handle cross-chain operations. NEAR Applications can get limited access keys without full account control to benefit from enhanced context and behavior. 

• Data Labelling: Data labelling refers to correctly uploading and validating data that can eventually be used for model training. Teams like PublicAI, Human Protocol, and NEAR Crowd have historically collected vast sums of human data for model improvement. With data labelling, the NEAR blockchain functions as the coordination layer for monetizing data contributions from contributors around the world. Data labelling ensures that all data examples used to train a language model are accurate and useful. Consistent model finetuning requires correctly labelled and cleaned data, such that model weights can be improved for specific tasks over time.

• Synthetic Data Generation and Optimization: Synthetic data sets are artificially created datasets that copy the patterns of real data. They are used to train AI in edge cases, or to develop capacity of LLM or SLM’s across data that would otherwise be difficult to utilize. Synthetic data is a method for taking LLM training beyond the limitations of the known world, and into broader and more complex fields where patterns of thinking and inference can benefit. In line with the goal of creating high performance models on NEAR, synthetic data generation is an important component for enabling NEAR native models to be competitive in the long-term. At the moment, teams like Fraction AI  are directly working on NEAR to accelerate synthetic data generation and optimization.

AI Agents, Agentic Protocols, and the Agent Interaction and Transaction Protocol (AITP):

The application layer of AI is built around AI Agents and Autonomous Agentic Protocols. NEAR’s vision for agents is multi-faceted: Assistants, agents, autonomous agents will use the Agent Interaction and Transaction Protocol (AITP) such that agents can exchange data and assets (leveraging NEAR Intents), for developing a robust agentic economy on NEAR. AI Agents on NEAR also refers to on-and-off ramps, agent hosting, audits for agents, and broad agentic frameworks. For the future of agentic commerce to be built on NEAR, these various agentic components must develop to work with each other. 

• Agent Interaction and Transaction Protocol (AITP): The Agent Interaction and Transaction Protocol (AITP) is a framework for enabling agents and agentic protocols to function with the capacity to communicate, interact with other agents, and transact value across chains. The table below indicates the development plans surrounding AITP: 

The ultimate goal of AITP is to become a Web3 native and multi-chain protocol for agentic payments, identification, human verification, and liveness. It is an ambitious project that will attempt to unify and standardize how all agents in the NEAR Ecosystem (web2 and web3 focused) will be able to connect and interact with one another. 

• Shade Agents and Agentic Protocols: Shade Agents and Agentic protocols represent the core Web3 native frameworks for deploying agents on NEAR via Phala Cloud using TEE-enabled GPUs. Shade Agents are multichain AI-powered smart contracts that are fully verified on-chain and designed to eliminate single points of failure. Using NEAR’s Chain Signatures for decentralized key management, Shade Agents leverage worker agents running in Trusted Execution Environments (TEEs) to securely access off-chain data and execute decisions using advanced AI models. Agentic Protocols, meanwhile, are multiple agents that work as a single entity capable of acting autonomously and in a financial capacity oftentimes with DAO based governance. Both Shade agents and Agentic Protocols are developed as NEAR native accounts that can leverage NEAR Intents and Chain Signatures to work across blockchains. Proximity Labs is currently running a Shade agent incubator program with close to a dozen shade agents and agentic protocols under development.

• PingPay: Pingpay operates as the on-ramp, off-ramp and multi-chain payment layer for the future of agents. Leveraging NEAR Intents, PingPay is positioned to offer automated subscriptions, pay widgets, card payments, and a native pingpay account with plans for creating a pingpay agent and API in the near future. As a homegrown project built by a NEAR native, PingPay remains under development. 

• Vijil: The auditor agent developed by Vijil is designed to improve the quality assurance and capacities of different shade agents or agentic protocols created on NEAR. The Vijil auditor agent will improve agent discovery by developing an auditor agent that performs security audits on agents on NEAR - across all frameworks - to prevent malicious agents from being called by any personal assistant agent. The Vijil Auditor Agent capabilities (among others) includes: 
• To block unauthorized signature access/usage such that the Agent consistently identifies and flags attempts to access or exploit NEAR signatures.
• The Agent successfully detects data exfiltration to external APIs and identifies attempts to send user data to unauthorized external endpoints.
• Issuance of certificates for secure agents: The auditor agent will automatically issue certificates for agents that pass security review.
• Registry maintenance of approved agents: The auditor agent will maintain an accessible, up-to-date list of all certified agents on NEAR.  

• Oauth with Privy: oAuth from Privy will enable agent developers to create chain-abstracted NEAR wallets when they sign up with a oAuth provider. The goal of this integration is to enable easy access to the NEAR account model (with all of its benefits) for non-crypto natives and developers, looking to leverage on-chain infrastructure for agent design. 

• Coinbase X402: As a launch partner in Coinbase's new agent monetization protocol, NEAR is aligned to grow its agentic economy in parallel with similar initiatives on Base and directly from Coinbase. The current focus of this partnership is on integrating liquidity from Base onto NEAR Intents, for agents to execute multi-chain transactions.  

• Agent Frameworks - Bitte Protocol and Eliza Labs: Agentic frameworks on NEAR enable developers and teams to rapidly prototype, develop and deploy custom agents into their application or for their own specific on-chain purpose. Two teams on NEAR currently have fully integrated agentic frameworks live: First, Bitte Protocol has developed a no-code agent builder, with the capacity to execute transactions using natural language prompts. Bitte is both a hub for multi-chain agents (including a wallet, registry, and marketplace of agents), and a platform for developers to deploy agents. Bitte is currently live on NEAR and can be leveraged as an agent builder for connecting agents to dapps on NEAR. Second, the Eliza framework is an AI-driven framework developed by Ai16z, for easy open-source agent development and character creation. Eliza agents are capable of being trained and designed with unique attributes and capabilities on-chain, with quick go-to-market time. 

• AI Assistants - Mia, Doppelgangers, and Unity: A number of early AI assistants have been embedded into NEAR native Dapps or integrated NEAR Accounts, including Mia (Sweat Wallet), Doppelgangers AI, the Unity Wallet Assistant, and upcoming assistants from VEAX and PingPay. While it’s still early for AI assistants, these assistants hold the promise of much more than operating as an advanced chat bot: With proper training and memory they become powerful primitives for predictive and personalized intelligence. 

• Agent Integrations with the NEAR House of Stake: Governance on NEAR has its own specific roadmap for implementing agents towards the governance process that has been outlined in three key phases: Phase 1) The AI Assistant, Phase 2) The AI delegate, and Phase 3) the AI CEO. As the roadmap below outlines, this phased approach will continue to develop over the coming months well into 2026.  

Conclusion - AI Alpha is NEAR: 

When NEAR is referenced as the ‘blockchain for AI’ what that actually means is that NEAR is the only open infrastructure provider in the world targeting all facets of the AI product stack (private compute, model training, monetization, and hosting, agents, agentic protocols, and assistants) on the world’s most scalable and decentralized coordination mechanism: The NEAR blockchain. With this framework, AI products and businesses on NEAR will be decentralized, private, verifiable, monetizable, and most importantly, user-owned. Your AI will work for you, and you alone on NEAR. 

This primer is an overview of the current state of the AI ecosystem on NEAR. It foreshadows a number of upcoming announcements, publications and releases across this stack that are confidential until ready for prime time. Stay on the lookout - these products will flex some of the best of NEAR’s DCML, Agentic Protocol, and AI Assistant capacities! 

Disclaimer:

The information contained in this primer is provided solely for general information. It does not constitute (and must not be construed as) (i) investment, financial, trading, or any other form of professional advice, (ii) legal or tax advice, or (iii) an offer to sell, a solicitation to buy, or a recommendation to acquire any token or other digital asset in any jurisdiction. Before making any decision to utilize a protocol or purchase tokens, you should conduct your own thorough research and due diligence, carefully review the project’s white-paper and other official materials, and consult independent professional advisors (legal, financial, tax and/or accounting) suited to your particular circumstances.