Why I built this

I'm Zeeshan. While writing emails, papers, blog posts, I constantly run into the same problem: I know the word exists, I can describe it exactly, but I just can't recall it. So I Google it, scroll through SEO-stuffed listicles, and maybe find it five minutes and some annoyance later.

I got tired of that loop. So, I built WordFor: a reverse dictionary that does one thing well: you describe a concept and it instantly gives you the word.

WordFor is best when you know a concept or feeling but not the word itself. Unlike a thesaurus (word → meaning), WordFor works in reverse (meaning → word). It runs entirely in your browser, has no accounts, no tracking, and no server

How it works

1

Describe

Type a description, definition, or a feeling in plain English.

2

Match

A tiny AI model converts your description into a meaning vector and compares it against 350,000+ dictionary definitions using 1-bit binary scoring. On desktop, a two-stage pipeline narrows candidates via fast Hamming distance, then reranks the top 500 with int3 precision. On mobile, pure binary scoring keeps downloads small (~30 MB).

3

Discover

Get ranked results instantly. Everything runs in your browser — no data ever leaves your device.

Examples

“unable to be put into words” → ineffable
“pleasure from someone else’s pain” → schadenfreude

Under the hood

WordFor uses a sentence-embedding AI model (mdbr-leaf-mt) running via Transformers.js to convert your description into a meaning vector, then compares it against 350,000+ pre-embedded dictionary definitions using a two-stage binary + int3 scoring pipeline. A lighter static embedding mode is available as fallback.

For a deep dive into the architecture, scoring pipeline, and evaluation results, see the blog post.

API access

WordFor runs entirely in the browser, so there's no traditional REST API. However, you can embed WordFor anywhere or link directly to a search:

URL query

Link to a pre-filled search: ?q=fear+of+heights. Add &mode=lite or &mode=full to force a mode.

Embed via iframe

Embed WordFor in your page:
<iframe src="https://wordfor.xyz/?q=your+query" width="100%" height="600"></iframe>

Licensing & attribution

WordFor’s code is licensed under CC BY-NC-ND 4.0. The dictionary data comes from multiple sources with different licenses:

Open English WordNet 2025+ — CC BY 4.0
Webster’s 1913 Dictionary — Public domain
GCIDE Webster 1913 portion — Public domain (GPL additions excluded)
Century Dictionary (1889–1911) — Public domain
Moby Thesaurus — Public domain
Wiktionary — CC BY-SA 3.0, sourced via kaikki.org (used at build time for quality signals and fine-tuning only; not redistributed)
ConceptNet 5.7 — CC BY-SA 4.0 (used at build time for quality signals only; not redistributed)

Privacy

WordFor has no server, no database, and no cookies. The AI model and dictionary data are downloaded once and cached in your browser. Every search happens locally on your device. Your queries are never sent anywhere. I use GoatCounter for privacy-friendly, cookie-free page-view analytics: no personal data is collected.