Catch New AI Search Terms the Day They Appear With new-term-radar

By the time a keyword tool lists "Seedance" or "Strata", the pages that rank for it are already written. new-term-radar is a 2k-star Python tool, shipped with three Agent Skills and an MCP server, that catches new AI names where they are said first and tells you which generic searches sit behind them.

Why This Skill Matters

The README lays out the trail a new name leaves, in order: a lab or well-followed account says it on X, or a model enters Hugging Face's trending list; one GitHub repo takes off; directories and competitors add pages and someone registers a domain; your own site gets its first impressions; and only then does Google Trends show the curve, days or weeks later. Keyword tools read that last step. The radar reads the earlier ones every day and uses the last one only to check itself.

The example it was built around: Niko1221/Strata was created on 2026-09-24 and had 7,700 stars nine days later. People were searching "niko strata" from September 26, while no keyword tool listed it.

Installation

Install the three skills with the skills CLI (the README passes -a codex or -a trae for agents other than Claude Code):

npx skills add zhuyansen/new-term-radar

In Claude Code they also install as a plugin:

/plugin marketplace add zhuyansen/new-term-radar
/plugin install new-term-radar@new-term-radar

The skills run the pinned release from PyPI (uvx new-term-radar==0.6.1), keep their state in ~/new-term-radar, tell you the cost and wait for your yes before any paid run, and treat what they read as data rather than instructions. The plugin also brings the MCP server and asks for your AIsa key, keeping it in the system keychain.

For the command line, Python 3.10 or newer with only the standard library:

git clone https://github.com/zhuyansen/new-term-radar
cd new-term-radar
cp config.example.json config.json      # then edit: your X accounts, sitemaps, topics
export AISA_API_KEY=...                 # from https://go.aisa.one/yansen-zhu; or `aisa login` with the AIsa CLI
python -m radar daily --out report.md

Everything the core run needs goes through one AIsa key — X posts, DataForSEO Trends, domain data, and the models that judge names. An optional second key, GOANYAPI_API_KEY, adds Google Trends related queries and one-country curves to the keyword table for about $0.07 a day. The README discloses that both links are referral links and that the radar works the same with a key from anywhere on aisa.one. Without a key, the free signals (Hugging Face, GitHub, sitemaps, Search Console) still run, but the report lists every term unsorted because nothing judges which are names.

Real Workflow: Find Tomorrow's Keyword Today

The three skills split the job:

1. Ask your agent for "today's new AI keywords". The first run only builds snapshots, so it reports GitHub breakouts alone — a repo under 14 days old averaging 200+ stars a day, or +400 since yesterday. From the second run on, every signal reports what changed.
2. Read the candidate table. Each term carries its signals, a name score from a small model (is this the name of one specific thing, or a common word?), and the search variants people use.
3. Read the keyword table under it. For each new name, the radar pulls the most-liked posts and a repo's README as evidence, has a model write the generic keyword an ordinary person would type for each job it sees, screens and sizes those phrases, and gives a decision: prioritize (rose in the last 30 days, still up, at least min_daily searches a day), watch (rose but smaller, too early to read, or flat with steady demand of that size), or skip (fading, or flat and small). A prioritize whose page one is crowded drops to watch.
4. Come back in three days. Each name is read twice, because on announcement day the posts are announcements and no use shows yet. The second reading is the one to trust.

With the example config (35 X accounts read three times a day, the keyword table for 5 names a day), the README puts a month at about $15 at AIsa's October 2026 prices.

Tips

• Put the terms you already cover in known_terms; a lead that contains one, or is contained in one, is marked and kept out of review.
• Read each site's robots.txt before adding its sitemap, and keep pause at or above its crawl-delay.
• The MCP server exposes three tools: recent_leads is free and instant, keyword_table costs about $0.05 a name plus $0.09 a call and takes 40 to 90 seconds a name, and run_radar is a paid run that takes minutes — the README says the schedule belongs in cron or Actions.
• The bundled GitHub Actions workflow runs sources every 8 hours and daily once a day, but stays off until you set the repository variable RADAR_ENABLED to true.
• Some sources carry terms: the README says TrustMRR data and Reddit posts are for your own research, not for republishing.

When Not to Use This

The README is candid about the limits. For pure model releases the evidence is announcements and benchmarks, and the proposed phrases came out weak; it works better for products and repositories that do one job. Hit rates per signal are "not yet" measured — the 14-day review has only begun, so treat any signal as unproven until your own report accumulates numbers. The X signal sees only names your account list says. And the built-in stop words are tuned for AI tooling; another field needs its own stop_words.

See the leaderboard for more skills.