NEAR's Brave AI rollout raises a hard question for the token

Brave has given NEAR AI something most crypto-linked AI projects struggle to show: a product in a widely used browser. On October 15, Brave said its Leo assistant began offering encrypted conversations with the GLM 5.3 Flash model running in NEAR AI's confidential-computing environment. A day later, NEAR AI introduced a live marketplace where customers can hire AI agents for individual jobs.

Those launches make the project more interesting. They do not, by themselves, answer the question facing anyone evaluating the NEAR cryptocurrency: when a browser user asks Leo a question, or a customer pays an agent, what economic activity actually reaches the token?

The answer is more complicated than the announcements suggest. NEAR has published mechanisms that can use or buy its token. But its current Agent Market payment guide describes customer and builder balances in dollars and USDC, and neither Brave nor NEAR AI disclosed the price, volume or settlement terms of the Leo arrangement. That gap between product adoption and token demand is the story investors need to watch.

What changed in Brave Leo

The Brave announcement says conversations with GLM 5.3 Flash are encrypted in the browser before they leave the device. Brave relays the encrypted request, while an attested trusted execution environment operated by NEAR AI decrypts and processes it. Brave says the gateway removes the user's IP address before forwarding the request to NEAR AI. The feature is available in Brave's Release, Beta and Nightly channels.

This is a specific model choice, not a blanket claim that every Leo conversation uses NEAR AI. It also extends a November 2025 experiment rather than creating the partnership from scratch. The October step matters because the private-inference option has moved beyond a test channel into the production browser.

Brave's technical account is unusually detailed. It describes hardware attestation, reproducible checks of the software images and an open-source gateway. But it also says the browser currently relies on Brave's gateway to verify the enclave before it receives an encryption key. Browser-side verification is a planned next step. The system's privacy claims are therefore more precise than a general promise that nobody can ever inspect an AI service; the currently deployed verification path still has a defined trust boundary.

NEAR AI describes the rollout as giving its private-computing service access to a browser with more than 128 million monthly active users. That number is Brave's overall reach, not a disclosed count of users choosing GLM 5.3 Flash, paying for Leo or generating NEAR AI inference revenue. The companies have not released those figures in the NEAR AI announcement.

The new Agent Market exposes the token question

On October 16, NEAR AI described Agent Market as live. Customers can hire an agent at a posted price or invite bids. Funds are held while the job is done. Once the buyer accepts the work, the builder receives the payment less a marketplace fee. An automated dispute process and builder reputation are meant to make small, repeatable jobs practical.

The distinction echoes a separate issue TECHi found in the FET agent-payment story: a service can be useful without making its associated token mandatory for every payment. Here, the current payment guide gives us a specific way to test that claim.

The product pitch is easy to understand: a working agent can sell a task without its builder creating a storefront, billing system and dispute desk. NEAR AI says a builder can run the agent on its own infrastructure; the marketplace passes a brief in and receives a result. One listed example, a research agent called Geo, charges $1 per call. Those are product details, not evidence that the marketplace has already reached material transaction volume.

The current payments and disputes guide says the platform takes 5% of each completed job, with no fee on refunds. It describes balances in USD and USDC. Builders withdraw dollars through Stripe Connect or USDC to a wallet, and the currencies do not convert automatically. The October launch article describes the same two payout routes.

That is a notable distinction from the February introduction, which said agents would be paid in NEAR. It does not prove the token plays no role behind the scenes or in other Agent Market functions. It does mean the current customer-facing documents do not establish that every job requires a buyer or builder to acquire NEAR. A dollar price for an AI task cannot simply be counted as token demand.

There are still unanswered operational questions. How many paid jobs have been completed, rather than merely listed? What share is paid in USD versus USDC? Where does the 5% fee go after collection? Does any portion fund NEAR purchases, network fees or token-denominated payouts? The public documents reviewed for this article do not provide those numbers.

Where NEAR already has an economic role

The token connection is not imaginary. In July, NEAR announced an AI staking payment option: users can stake NEAR and receive recurring compute credits for confidential inference and agents. The underlying tokens remain theirs and can be withdrawn after unstaking. If users choose that payment route at scale, it could tie AI usage to tokens locked for the duration of use.

But this is an available payment mechanism, not evidence that Brave's Leo traffic is funded through it. Brave did not say its users must stake NEAR to ask Leo a question. Likewise, Agent Market's current payment pages foreground USD and USDC. Treating every private AI prompt or agent hire as an onchain NEAR transaction would overstate what the companies have disclosed.

There is a separate, established token path in NEAR Intents. In a February tokenomics explanation, NEAR said its Intents fee switch was live, revenue-share payouts to distribution partners were denominated in NEAR, and Intents revenue had been directed to more than 1 million NEAR in buybacks. The same explanation discusses a governance-based structure that could direct future product revenue toward buybacks, staking or other supply measures, while explicitly saying no specific future use is guaranteed.

That matters for the Brave and Agent Market news because it shows how an AI product could be connected to token economics. It does not show that the 5% Agent Market fee is already buying NEAR, or that Brave usage is being settled through Intents. Those links need transaction records, fee statements or other specific disclosure.

The measure that would separate adoption from promotion

The first useful metric would be actual Leo usage of the NEAR-hosted model: requests, active users or paid inference volume over a defined period. Brave's monthly user figure is a ceiling on potential reach, not a conversion rate. Even a large number of requests would tell only the first part of the story. The second part is whether the associated revenue is earned by NEAR AI, passed to the protocol, used to purchase NEAR, or retained elsewhere.

For Agent Market, the comparable figures are completed paid jobs and gross payment volume, broken out by currency. As TECHi noted when Coinbase let agents pay merchants, payment capability is different from proof of merchant adoption. A marketplace can show many agents and still process little economic activity. Its 5% fee becomes meaningful only when there is repeat demand, and token holders cannot assess value capture without knowing how the fee is distributed.

Finally, the project could publish a reconciliation between AI activity and onchain token flows. That would let readers distinguish three different claims: people are using a product built by a NEAR-linked company; customers are using the NEAR blockchain; and product revenue is creating demand for the NEAR token. The claims may overlap, but one does not automatically prove the next.

There is a product risk, too. Brave's privacy design depends on a particular model and verification pipeline. Users may prefer a different model, and Brave itself says browser-side attestation checking has not yet shipped. A promising security architecture still has to win usage, retain users and survive independent scrutiny. The announcements provide a basis for measuring that progress; they do not substitute for it.

NEAR now has something more substantial to discuss than an AI roadmap: a live browser integration and a marketplace with documented payments. The strongest case for the token will come when the companies show how much work those products do and exactly which part of that work, if any, creates durable NEAR demand. Until then, the launch is a real product event with an economic question still open.