NEAR AI

NEAR AI is building user-owned AI: systems where people and organizations can use powerful models and agents while retaining control of their data, credentials, identity, and digital activity.

Founded by Illia Polosukhin, co-author of Attention Is All You Need, NEAR AI is developing the infrastructure for private inference, secure agents, user-owned data, and agent-to-agent coordination.

Private and verifiable inference

NEAR AI Cloud provides OpenAI-compatible access to leading open-weight models running inside confidential hardware.

Sensitive prompts and outputs remain isolated inside hardware-attested environments, while cryptographic verification allows developers to confirm the model, code, and execution environment instead of relying only on provider promises.

Explore models · Read the docs · Verify inference · Service status

Secure agents with IronClaw

IronClaw is an open-source, security-first agent harness built for confidential computing.

It separates reasoning from action, keeps credentials outside the model context, isolates tools, scopes permissions, and records an auditable history of agent activity. The goal is to let agents safely work across applications, services, and financial systems on behalf of users and organizations.

User-owned data with Trace Commons

Trace Commons gives users control over the records created by their agents: objectives, decisions, tool calls, workflows, outcomes, and evidence of completion.

Instead of disappearing into proprietary platforms, these traces can remain private, be selectively shared, improve future models, or create value for the people and organizations that produced them.

The full stack

Private inference protects what an agent thinks. IronClaw protects how it acts. Private model routing selects the right model for each task without leaking sensitive traces. Trace preserves ownership of the work agents produce. NEAR provides neutral infrastructure for identity, permissions, payments, verification, settlement, and coordination.

Together, these systems form the foundation for an agent economy where users own their AI, their data, and their economic agency.

Research

NEAR AI research explores how confidential computing, cryptographic verification, and decentralized infrastructure can support private, user-owned AI.

Proof of Response

A framework for proving that decentralized systems and agents produced a specific response within a verifiable execution window.

Decentralized Confidential Machine Learning

A technical vision for training and running AI across distributed infrastructure while preserving confidentiality and verifiability.

Benchmarks (https://nearai.github.io/benchmarks/)

NEAR AI publishes reproducible evaluations of models and agents across practical tasks including coding, tool use, document workflows, and multi-step execution. Compare model and agent performance across public benchmarks, inspect individual runs, and see how different implementations perform on real-world tasks.

Explore NEAR AI on GitHub

Browse the open-source infrastructure, agent systems, verification tools, and research projects behind NEAR AI.