Stop guessing icon names. Describe what you need ("sad robot," "money leaving") and find it instantly across 200K+ icons from Lucide, Heroicons, Phosphor, Material and 80 more sets. Copy as SVG, React, Vue or CSS.
What's great
the semantic search actually works, not just on paper. I searched "money leaving" and "phone ringing but nobody picks up" and both returned the right icons (moneybag-move-back, cash-move, phone-missed) despite neither query matching an icon name. no signup, copy as SVG/React/Vue/CSS in one click, and it pulls from 80+ icon sets so you're not stuck re-searching across Lucide vs Heroicons vs Phosphor separately.
What needs improvement
the license info could be surfaced closer to the icon itself. right now you have to check which set an icon came from and separately know that set's license, an inline badge per result (MIT, Apache, CC-BY etc) would save a trip to each icon set's own page before using one commercially.
vs Alternatives
I'd normally just google "heroicons trash icon" or dig through Iconify's site directly. those work fine if you already know the set and rough name, but IconsDB skips the "which set has this" step entirely since it searches all of them from one box.
Report
3 views
tried "money leaving" and "phone ringing but nobody picks up" and both actually surfaced the right icons (moneybag-move-back, cash-move, phone-missed) despite neither query being a literal icon name. that's the actual hard part of semantic search, most tools that claim it just do fuzzy string matching on tags underneath. is the matching purely on icon names/tags/aliases, or did you also embed something like the icon shape itself, since a few duotone variants showed up that don't obviously share tags with the outline ones?
@galdayan currently it only works on the embeddings on top of icon names. We are working on analyzing every icon and embedding it's visual identity as well, will be out soon.
makes sense, name/tag embeddings alone already got both my test queries right so that's a solid foundation. visual embedding sounds like it'd mainly help with icons that are mistagged or under-tagged rather than fix a core gap. curious if you'll expose a toggle to search by visual similarity separately once it ships, or just blend it into one ranked result set
Been testing this internally for a few days now, crazy accurate results