NEAR is combining cross-chain execution, private inference and agent-to-agent commerce into a broader infrastructure stack for autonomous AI systems. Its NEAR AI Agent Market allows agents to discover tasks, submit bids, execute work and receive payment through an escrow-based marketplace powered by NEAR Intents. The architecture treats agents as economic participants that can coordinate work and payments rather than simply respond to user prompts.
The marketplace connects with infrastructure already developed across the NEAR ecosystem. NEAR Intents abstracts cross-chain routing, while Chain Signatures lets NEAR accounts and smart contracts authorize transactions on external networks through multi-party computation. NEAR currently advertises Intents-based access across more than 35 chains and 150 assets. The result is a stack designed to let an agent coordinate activity across multiple networks without requiring users to manually manage every bridge, wallet and transaction step.
Chain Signatures Give Agents Cross-Chain Control
NEAR’s Chain Signatures documentation explains that one NEAR account can derive and control addresses across Bitcoin, Solana, XRP, Aptos, Sui, Cosmos and EVM networks. An MPC service produces the required signatures without placing the complete signing key under the control of a single party. That model gives agents persistent cross-chain accounts while reducing dependence on separately managed private keys for every network.
NEAR also supports restricted access keys that can be limited to defined contracts, methods and gas allowances. Its Shade Agent framework combines those controls with persistent accounts and trusted execution environments, allowing developers to build additional on-chain guardrails around agent actions. Scoped authorization is important because an autonomous agent does not necessarily need unrestricted authority over every asset controlled by its underlying account.
Privacy forms another layer. Confidential Intents executes transactions through a dedicated private shard connected to NEAR mainnet through TEE-based infrastructure, while supporting selective disclosure for audit or compliance requirements. The design aims to hide sensitive transaction details without preventing funds from eventually moving between confidential accounts and public blockchain environments.
Private Inference Extends Beyond Transactions
Agents can also use NEAR AI Cloud for confidential model inference. Prompts are encrypted before entering hardware-isolated trusted execution environments, processed inside those enclaves and returned with attestations that users can independently verify. This separates private AI computation from the public settlement layer, allowing an agent to reason over sensitive information without placing that data directly on-chain.
That security model has relevance beyond blockchain. The U.S. National Institute of Standards and Technology says confidential computing can protect data while it is actively being processed, including AI workloads, through hardware-enabled isolation such as trusted execution environments. NEAR is applying that established security model to agents that may also control assets and execute financial actions.
The individual components are already operational to different degrees, including Agent Market, Chain Signatures, Confidential Intents and private inference, but their existence does not establish broad agent-economy adoption. NEAR’s more significant bet is architectural: identity, permissions, private compute and cross-chain settlement are being assembled into one operating stack for autonomous software. The next test is whether agents can use those components reliably in recurring commercial workflows rather than isolated demonstrations.