new-term-radar: catch names like "Strata" the day they appear, long before keyword tools notice
What this project does
new-term-radar is a daily radar built to catch AI names that are just starting to surge: new models, new products, new tools, and new repositories. Its core pitch is not "organize words that are already hot" but "catch them before any keyword tool has indexed them." It is aimed at people who build pages, posts, or products around new terms — SEO for AI tool sites, newsletter authors, directory owners, anyone who wants to know about "Strata" or "Seedance" on the day it appears rather than two weeks later. Put simply, it turns the information gap into a subscribeable service you can run every morning.
Why it is trending
Anyone publishing AI content shares one pain: by the time Google Trends draws a curve and keyword tools index a term, you are already late. This project grabs exactly that time gap. The author makes the point with a real example: the repo Niko1221/Strata was created on 2026-09-24 and had 7,700 stars nine days later, while people were already searching "niko strata" from 9-26 — yet at that moment no second repo carried the name and no keyword tool listed it. A single repo gaining 800 stars a day is, by itself, a public and strong signal. That practical value — know a term a day earlier, claim a search slot a day earlier — pushed the project past 1,000 stars in a week.
Technical highlights
It reads six signals and then hands them to a "name judge" that decides what is worth reporting. Signal one is X: names used for the first time in 60 days by accounts on your list. Signal two is new models entering the Hugging Face trending list. Signal three is GitHub repos that suddenly take off on their own. Signal four is sitemaps and directory sites. Signal five is Reddit "ask" threads where people vent about a problem. Signal six is Google Trends, used only to cross-check whether the earlier catches were right. Every day it produces a report, and it runs a 14-day revisit check to see which flagged names actually blew up. It also ships three ways to consume it: a command line, three Agent Skills, and an MCP server, so you can drop it straight into your own agent workflow.
Who it is for
I think it fits three crowds best. First, SEO for AI tool and directory sites, who can build a page for a new term before competitors do and harvest early search traffic. Second, newsletter and social-media authors, who can turn "what surfaced today" into a topic — readers naturally love "you knew before others did." Third, indie developers, who can use it to judge whether a direction is genuinely hot and avoid wasting two weeks on a fake wave. At its core it is an "early-trend smell amplifier."
Quick start
Install is one line: pip install new-term-radar. After that you can run the daily radar from the command line, or wire its three Agent Skills or its MCP server into tools like Cursor or Claude Code so it pushes new terms to you automatically each day. The author also ships a pains command that specifically grabs Reddit signals about "what problem is everyone suffering from."
How it compares
Against keyword tools like Google Trends or Ahrefs, new-term-radar reads the earlier stage — X posts, HF trends, a single repo's spike — instead of waiting for search volume to form. Against a bare GitHub Trending subscription, it normalizes multiple sources into one "name judge" that cuts noise. In short: keyword tools tell you "it is already hot," this tells you "it is about to be hot." For people who live on the information gap, that half-day-to-two-week head start is real money.
One more thought
The author is also refreshingly honest about the cost of each signal. Reading X needs an AIsa key and runs about 0.003 dollars per account per run; Hugging Face trends, GitHub spikes, sitemaps, and Reddit each carry their own cost. So it is not free magic, and a heavy schedule will show up on a bill, but the unit price is low enough for an individual. Another point to watch is false positives: a name mentioned once by one big account does not mean it will blow up, which is exactly why the "name judge" step and the 14-day revisit check are the real value — it does not promise every term lands, it compresses the noise and pushes the genuinely rising signals to the front. If you wire it into an agent workflow, I would treat the report as a "candidate pool" rather than a "must-do list," and glance at it by hand before acting, so the value of the information gap is not diluted by junk terms.