About NEARPulse AI

One question, tested with real NEAR data: is mainnet usage broadening, or is it just a few busy days? Everything on this site traces back to a SQL query you can run yourself.

What the data says right now

Historical snapshot, 2026-08-06 to 2026-11-04. Not live data.

Latest 7 days against the 7 before: transactions +3.4% (577,201 to 596,965 per day), distinct signers +7.1% (51,827 to 55,529 per day). Both rises lean on 2026-11-03 and 2026-11-04, and the cause of that jump is not verified.
Latest 30 days against the 30 before: transactions -17.7%, distinct signers -13.4%.
On 10, 11 and 15 October, a handful of receiver accounts took most function-call actions while the number of distinct contracts did not rise.

Verdict so far: mixed. Neither confirmed nor rejected. The full write-up is on the Research Reports page.

Check it yourself in two minutes

Open Data Explorer and copy one of the two SQL queries.
Paste it into the BigQuery console (the free sandbox is enough), read the "This query will process" estimate, then run it.
Compare the rows with the CSV files in the data folder of the GitHub repository.
Ask Ask NEAR AI which day had the highest signers. The answer should be 2026-10-04 with 177,196.

What this cannot tell you

Distinct signers is not unique users and not all active accounts. One person can control many accounts, and one account can be a bot.
Function-call actions are actions, not transactions. One transaction can create several.
Only 91 days are covered, and the data is a snapshot, not a live feed.
The causes of the October and early-November jumps were not investigated, so nothing here explains them.
Nothing here is about token price, and it is not investment advice.

How Ask NEAR AI stays honest

The model only receives the verified metrics stored in this app.
After it answers, the app checks that every number in the answer exists in those metrics. If one does not, the answer is dropped and a built-in explanation is shown instead.
Price questions never reach the model.
Each answer is labelled with the mode that produced it. The model can still word things imperfectly, so check important numbers against the report.

Built with

Data from the public NEAR dataset in Google BigQuery. Next.js, TypeScript, Tailwind CSS, Recharts and Vitest. Answers come from an OpenAI-compatible model on NEAR AI Cloud. Built by one person with an AI coding assistant (Claude by Anthropic); the queries were run and the results checked by the author. Built for NEAR Hackathon 2026, Investment Thesis track.

Code and thesis: github.com/cryptobyrj/Nearpulse-ai