Font Awesome

This explorer contains 3,860 icons from Font Awesome Pro 6.5.1 and Free 7.2.0, spanning 10 style families across 3 icon families (classic, sharp, duotone). The pro collection includes 1,889 pro-exclusive icons and 1,971 free-tier icons with 550 brand logos. Each non-brand icon is available in up to 9 style variants: solid, regular, light, thin, duotone, sharp solid, sharp regular, sharp light, and sharp thin.

Search uses a hybrid strategy: FTS5 keyword search runs on every keystroke at 0.18 ms average latency — 1,151x faster than embedding search — catching exact and prefix matches like "cat" → cat, cat-space, cauldron. When fewer than 10 keyword results are found, semantic embedding search (all-MiniLM-L6-v2, 384 dimensions) kicks in at ~35–55 ms in-browser, finding conceptual matches: "furry pet" → dog, cat, paw, paw-claws, hippo; "spinning loader" → spinner, spinner-third, circle-notch (80% precision); "navigation menu" → bars, ellipsis, caret-down. Score fusion blends both: 0.7 × FTS5 + 0.3 × cosine.

Press / then type a keyword, name, or concept to search. Smart mode is enabled by default — it combines keyword/prefix matching with semantic embedding search for the most comprehensive results. Click any icon for details and copy formats.

/ Focus search

Color

Duo

Style Comparison

Color Duo

Copy Formats

Metadata

About

Font Awesome Icon Explorer

Repository

github.com/tankbottoms/font-awesome v1.0.0

Icon Packages

Pro 6.5.1	3,762 icons — Released 2023
Free 7.2.0	1,970 icons — Released 2025
Free 6.7.2	1,895 icons — Released 2024

Combined Inventory

Total	3,860 unique icons
Pro-only	1,889 icons
Free	1,421 icons
Brands	550 logos
Styles	10 (solid, regular, light, thin, duotone, sharp x4, brands)
Families	3 (classic, sharp, duotone)

Explorer

Built	2026-03-15
Stack	Bun, TypeScript, SQLite FTS5, sqlite-vec, transformers.js
Search	FTS5 (keyword) + all-MiniLM-L6-v2 ONNX (semantic)
Deploy	Cloudflare Workers

Design Intent

This explorer is a reference template for building minimal, fast, in-browser search tools. The architecture demonstrates hybrid search: FTS5 keyword search for instant prefix/exact matching at sub-millisecond latency, combined with semantic embedding search (all-MiniLM-L6-v2 ONNX) for conceptual queries -- all running entirely client-side with no backend API required for search. The single-file approach (inline CSS, JS, and data) keeps deployment trivial while the neo-brutalist design system keeps the UI focused on content over chrome.

Benchmark

View full benchmark report (20 queries, FTS5 vs Embedding)

Benchmark Report

FTS5 vs Embedding Search -- 20 Queries

Executive Summary

Icons	3,860 \| Styles: 30,112 \| Embeddings: 3,860
Database	30.32 MB
Model	all-MiniLM-L6-v2 (384 dims, normalized)

Method	Avg Latency	Avg P@5	Best For
FTS5 (keyword)	0.181 ms	37.8%	Exact matches, unicode, prefix
Embedding	208.6 ms	32.8%	Conceptual, descriptive queries
Hybrid	< 1 ms typical	Best of both	Production strategy

Key Findings

FTS5 is 1,151x faster than embedding search (0.18ms vs 209ms)
Embedding wins on conceptual (40% vs 31%) and description (80% vs 20%) queries
FTS5 wins on exact (50% vs 20%) and unicode (100% vs 0%) queries
11 of 20 queries tied -- both methods find similar results
Embedding wins 6/20, FTS5 wins 3/20 (excluding ties)

Precision by Query Type

Type	Queries	FTS5 P@5	Emb P@5	Winner
exact	2	50%	20%	FTS5
conceptual	7	31%	40%	Embedding
synonym	8	27%	32%	Embedding
description	1	20%	80%	Embedding
unicode	2	100%	0%	FTS5

Notable Embedding Wins

Query	FTS5 Top	Embedding Top	Verdict
"spinning loader"	loader (1 result)	loader, spinner, spinner-scale, spinner-third, circle-notch	Emb finds 4x more
"navigation menu"	compass-slash, compass, bars	bars, ellipsis, ellipsis-vertical, caret-down	Emb nails hamburger
"play video"	circle-play, play, file-video	play, circle-play, youtube, vimeo, file-video	Emb finds platforms
"notification alert"	bells, bell-exclamation	bell-on, square-exclamation, triangle-exclamation	Emb finds alert variants
"lock security"	lock, lock-keyhole	lock-open, lock, unlock, lock-keyhole	Emb finds unlock too

All 20 Queries -- Detailed Results

Query	Type	FTS5 P@5	Emb P@5	FTS5 Lat.	Emb Lat.
"arrow"	exact	80%	20%	0.71ms	6.37ms
"house"	exact	20%	40%	0.41ms	5.92ms
"social media"	conceptual	20%	20%	0.33ms	6.30ms
"money"	synonym	0%	20%	0.31ms	5.52ms
"navigation menu"	conceptual	60%	60%	0.22ms	6.46ms
"cloud computing"	conceptual	0%	40%	0.18ms	6.29ms
"spinning loader"	description	20%	80%	0.09ms	6.22ms
"delete remove"	synonym	40%	0%	0.25ms	6.16ms
"notification alert"	conceptual	40%	60%	0.31ms	6.57ms
"play video"	synonym	60%	20%	0.41ms	6.07ms
"user profile"	synonym	0%	20%	0.53ms	6.01ms
"download file"	synonym	75%	50%	0.37ms	5.86ms
"chart analytics"	conceptual	60%	60%	0.22ms	6.20ms
"lock security"	conceptual	20%	20%	0.32ms	6.38ms
"f015"	unicode	100%	100%	0.22ms	0.22ms
"f007"	unicode	100%	100%	0.21ms	0.22ms
"calendar date"	synonym	0%	40%	0.17ms	6.23ms
"email message"	conceptual	20%	20%	0.24ms	6.58ms
"gear settings"	synonym	40%	20%	0.18ms	6.27ms
"heart love"	synonym	0%	25%	0.20ms	6.44ms

Database Size Breakdown

Total database	30.32 MB
Icons + FTS5 index	~24.66 MB
Embeddings (3860 x 384 x float32)	~5.65 MB

Recommendations

1. FTS5 is the primary search method:
   Sub-millisecond latency, excellent for exact/prefix matches

2. Embedding search excels at conceptual queries:
   "social media" finds brand icons, "spinning loader" finds spinner variants

3. Hybrid strategy:
   Run FTS5 first on every keystroke; if < 10 results, trigger embedding fallback

4. Score fusion:
   final = 0.7 * fts5_normalized + 0.3 * cosine_similarity

5. Unicode shortcut:
   Direct lookup when query matches /^[0-9a-f]{4,5}$/i

Documentation

Build Pipeline, Search System & Usage

Quick Usage

Solid:    <i class="fa-solid fa-house"></i>
Regular:  <i class="fa-regular fa-house"></i>
Light:    <i class="fa-light fa-house"></i>
Thin:     <i class="fa-thin fa-house"></i>
Duotone:  <i class="fa-duotone fa-house"></i>
Sharp:    <i class="fa-sharp fa-solid fa-house"></i>
Brands:   <i class="fa-brands fa-github"></i>

Data Pipeline

Primary source	`icon-families.json` (Pro 6.5.1)
Supplement	`icon-families.json` (Free 7.2.0 — 98 extra icons)
Categories	`categories.yml` — 68 categories mapped to icons
Ingest script	`bun run src/build/ingest.ts`
Database	`data/icons.db` — 30.32 MB (SQLite + FTS5 + embeddings)

Database Schema

icons.db
├─ icons (3,860 rows)
│  name, label, unicode,
│  search_terms, aliases,
│  categories, tier
├─ icon_styles (30,112 rows)
│  family, style, svg_path,
│  svg_viewbox, width, height
├─ icons_fts (FTS5 virtual)
│  name, label, search_terms,
│  aliases, categories
├─ icon_embeddings (3,860)
│  384-dim float32 blobs
│  all-MiniLM-L6-v2
└─ sources (2 rows)
   pro-6.5.1, free-7.2.0

Search: FTS5 (Keyword)

Engine	SQLite FTS5 full-text search
Avg latency	0.18 ms
Precision@5	37.8% across 20 benchmark queries
Best for	Exact matches, prefix search, unicode lookup
Exact queries	50% P@5 — "arrow" finds arrow-right, arrow-left, arrow-down
Unicode	100% P@5 — "f015" finds house, "f007" finds user

Search: Embeddings (Semantic)

Model	all-MiniLM-L6-v2 (384 dims, ONNX quantized ~23 MB)
Embedding script	`bun run src/build/embed.ts` — 8.9s for 3,860 icons
Avg latency	~35-55 ms (WASM in-browser estimate)
Precision@5	32.8% avg, but 80% on description queries
Best for	Conceptual & descriptive queries

"spinning loader" → loader, spinner, spinner-third, circle-notch "navigation menu" → bars, ellipsis, caret-down "social media" → microblog, reddit, tumblr, bluesky

Benchmark Summary

Queries tested	20 (exact, synonym, conceptual, description, unicode)
Embedding wins	6/20 queries
FTS5 wins	3/20 queries
Tied	11/20 queries
Speed ratio	FTS5 is 1,151x faster (0.18 ms vs 209 ms)
Full report	`data/benchmark-report.md`

Hybrid Strategy

Every keystroke triggers FTS5 first. When keyword results are sparse, semantic embeddings fill the gap.

1. FTS5 keyword search
   <1ms, runs on every keystroke

2. Embedding fallback
   Triggers when FTS5 < 10 results

3. Score fusion
   0.7 × FTS5 + 0.3 × cosine

4. Unicode shortcut
   Direct lookup for /^[0-9a-f]{4,5}$/i

Build Stack

Runtime	Bun 1.3.10 + TypeScript
Database	bun:sqlite (FTS5 built-in)
Embeddings	sentence-transformers / all-MiniLM-L6-v2
Client search	sql.js (WASM) + ONNX Runtime (planned)
Deployment	Cloudflare Workers
Design system	Neo-brutalist (monospace, rectangular, hard-offset shadows)