Free · private · no account
Group thousands of keywords by meaning, in your browser
Paste a list or drop an export from Ahrefs, Semrush or Keyword Planner. Get topic clusters you can build pages around, with intent labels and a CSV out. The model runs in this tab. Nothing is uploaded.
- 22 MB model, downloaded once
- 10k keywords in about a minute
- 0 servers, cookies or accounts
1Keywords
0 keywords · Volumes rank the clusters and pick their names. Up to 25,000 keywords per run.
2Settings
3Clusters
Intent labels are a rule-based guess from the words in each keyword, not from search results.
How it works
Each keyword becomes a vector
A small sentence-embedding model, the kind behind semantic search, turns every keyword into 384 numbers. Keywords that mean the same thing point the same way, even when they share no words.
Close vectors become clusters
The list is walked from highest volume down. Each keyword joins the nearest existing cluster if it is similar enough, otherwise it starts one. Then every keyword is rechecked against every cluster until nothing moves.
Clusters become topics
The same pass runs again on the clusters themselves at a looser setting. The result is a two-level map: topics for hubs, clusters for pages, keywords for headings.
The tightness slider is the similarity a keyword needs to join a cluster. Around 0.55 you get broad topics, around 0.75 near-duplicates. Most lists look right between 0.58 and 0.68. Move it after a run and the page re-clusters instantly, because the vectors are kept.
How it compares
Honest version. The paid tools cluster by overlap in Google's results, which is the industry's definition of "same page". This tool clusters by meaning, which is free and fast and catches most of the same groupings, but not all.
| Keyword Clusters | Typical paid tool | |
|---|---|---|
| Price | Free | $50 to $60 a month |
| Where your keywords go | Nowhere. They stay in the tab | Their servers |
| Clustering method | Meaning (embeddings) | Search-result overlap |
| Topics above clusters | Yes | Sometimes |
| Intent labels | Rule-based guess | Usually from results |
| Keywords per run | 25,000 | Credits |
| Account needed | No | Yes |
| Export | CSV, Sheets, text | CSV |
Questions
Is anything uploaded?
No. There is no server behind this page. The model and the page are static files, and your keywords never leave the browser. The only third-party request is the open-source runtime library from a CDN. Open your browser's network tab and watch.
Which model is it?
all-MiniLM-L6-v2, an open sentence-embedding model, quantized to 8-bit so it is 22 MB. It runs through transformers.js on WebAssembly, using several CPU threads when the browser allows it.
How is this different from SERP clustering?
SERP clustering says two keywords belong together when Google shows mostly the same pages for both. That needs a search-results feed for every keyword, which costs money. This tool says two keywords belong together when they mean the same thing. On most lists the two methods agree. They disagree on keywords Google treats as one query despite different wording, and on phrases that read alike but get different results.
How big a list can I run?
Up to 25,000 keywords per run. Around 10,000 takes about a minute on a recent laptop, longer on a phone. The slider re-clusters instantly after the first run.
Can I use this for client work?
Yes. There is no license to worry about and nothing is stored. Download the CSV or copy for Sheets, and keep your own copy.
Why is it free?
It costs nothing to run, so it can cost nothing to use. If enough people want more, a paid version with saved projects, bigger runs and SERP overlap using your own API key is the plan. Leave an email below if that is you.
Want more from it?
Leave an email and tell us what you would need. One message when something ships. No list, no drip.