The agent economy, on NEAR

The operating system for agents

nearos gives every agent an account, a private place to think, and a way to act across chains. One runtime. The user keeps the keys.

What is nearos?

NEAR is the open infrastructure for the agent economy. nearos is the harness on top of it: an account for the agent, a confidential boundary for inference, and settlement that stays onchain.

Execution can stay private. The result does not. Assets and intelligence stay with the people who run the agents.

The runtime

Four layers. One agent.

01

An agent is an account.

Keys, balances, memory, and permissions stay with the agent. There is no shared wallet and no borrowed identity. The person who runs the agent holds the account.

02

Say the outcome. The runtime finds the path.

Intents are how an agent acts across chains. The same rail NEAR Intents already clears: sub-cent fees, sub-second settlement, no manual bridge.

03

Private by construction.

Prompts and tools run inside a hardware enclave. Credentials stay in the sandbox. The model does not receive the secret.

04

Finality you can wait on.

Actions settle on NEAR Protocol. Sharded execution, 600ms blocks, 1.2s finality. What the agent did becomes an onchain fact.

“An agent economy only works if the agents can hold value, act in private, and settle in public.”

nearos, on NEAR

The console

One surface for every agent you run.

nearos

Confidential mode

Account
ops.agent.near
Treasury
12,400 USDC · cap on
Intent
Settle across chains
Inference
Private enclave
Policy
Spend without a new signature

Last settlement

1.2s

Finality on NEAR

  • Identity

    The name, keys, and permissions the agent carries.

  • Treasury

    Balances the agent can spend, and the cap it cannot cross.

  • Intents

    The outcome it is working toward, across chains.

  • Inference

    Private prompts and tools, sealed in an enclave.

  • Policy

    What the agent is allowed to do without asking again.

  • Ledger

    What already settled, and when it became final.