zhuyansen/new-term-radar

The problem

If you run an AI tool site, a directory or a newsletter, the value of a new name is front-loaded: the day a model or repo appears is when a page about it can rank. The usual tooling — Google Trends, keyword planners — is backward-looking by design; it confirms demand after directories and competitors have already moved in. new-term-radar tries to shorten that gap by scanning the places a name appears first and reporting it the same day, with Google Trends used afterward only as a scorecard.

What's different

Keyword tools are built on search-volume history. This project inverts the pipeline: it treats a young GitHub repo gaining stars fast, a model entering Hugging Face trending, or new shared words across a directory's fresh URLs as the signal itself, and treats search data as validation. The notable piece is the keyword table: for each new name it derives the generic task-oriented phrase people would actually type, checks volume, and adds a page-one competition check, so you get a decision (prioritize / watch / skip) rather than a raw term list. A second differentiator is delivery surface — CLI, three Agent Skills and an MCP server — so the report can be consumed by an agent workflow, not just a dashboard. Its own 14-day Trends backtest is a reasonable honesty signal, though it is self-reported.

Who it's for

People who monetize being early on AI terms: SEO operators on AI tool directories, newsletter writers, and small product teams watching for naming opportunities. It is a poor fit if you need historical search-volume analytics, broad market research, or anything enterprise-grade; it is also a poor fit if you are not prepared to spend money per run on third-party API keys.

Cost of adoption

Python 3.10+, installable from PyPI (pip install new-term-radar appears in the README's quick start). Several signals require an AIsa key, and the README's own example configuration comes to roughly $15/month; GitHub, Hugging Face and sitemap signals are free, and Search Console needs your own Google token. The README is unusually detailed about costs and methodology, which is a good sign, but it is also the main source of claims — there are no third-party benchmarks. License is MIT. The project is days old, so expect API drift and breaking changes.

Verdict

The idea is sound and the execution is more thoughtful than most SEO scrapers: the name-versus-generic-keyword distinction and the page-one check are the parts that would actually save time. The project is brand new, not established — about 12 days old — and the star velocity it advertises (700+ on day one, 1,000 within three) sits alongside a fast-rising Python board position rather than any long track record or Hacker News discussion. That pattern is consistent with a launch spike; whether the daily reports stay useful after the novelty fades is the open question. Worth a trial run if you already pay for keyword data and want earlier leads, not worth restructuring a workflow around yet.