ua-lite

A compact, dependency-free, MIT-licensed user agent parser for JavaScript. Detects browser, engine, OS, CPU, and device — plus first-class classification of bots, AI training crawlers, and in-app browsers, and support for User-Agent Client Hints.

No build step. src/ua-lite.js is the distributable — drop it straight into a <script> tag, require() it in Node, or import it.

Install

npm install ua-lite

Or drop the file directly into a page:

<script src="ua-lite.js"></script>
<script>
  const result = UALite.parse(navigator.userAgent);
</script>

Quick start

const { parse } = require('ua-lite');

const result = parse('Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36');
console.log(result.browser.name); // 'Chrome'

Full example output

parse('Mozilla/5.0 (Linux; Android 14; Pixel 8) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.6367.82 Mobile Safari/537.36');

{
  "ua": "Mozilla/5.0 (Linux; Android 14; Pixel 8) AppleWebKit/537.36 ...",
  "browser": { "name": "Chrome", "version": "124.0.6367.82" },
  "engine":  { "name": "Blink", "version": "124.0.6367.82" },
  "os":      { "name": "Android", "version": "14" },
  "cpu":     { "architecture": null },
  "device":  { "type": "mobile", "vendor": "Google", "model": "Pixel 8" },
  "agent":   { "type": "browser", "name": "Chrome" }
}

parse() never throws and never returns undefined fields — every key is present, with null where nothing was detected.

API reference

parse(ua, hints?)

Synchronously parses a UA string. ua is trimmed to 500 characters. Pass an optional hints object (see Client hints) to merge Client Hints data into the result in one call.

parseAsync(ua?)

Returns a Promise<UAResult>. In a browser, automatically collects navigator.userAgentData (including high-entropy values like platformVersion, architecture, and model) and merges them into the result. Falls back to a plain parse() when the Client Hints API isn't available. ua defaults to navigator.userAgent.

resolveHints(hints, result)

Merges a Client Hints header object into an existing UAResult, returning a new object — the input result is never mutated.

collectHints()

Browser-only. Reads navigator.userAgentData synchronously and returns a hints object (low-entropy only — brands, mobile flag, platform). Returns null if the API isn't available.

Client hints

Browser-side

const { parse, collectHints } = require('ua-lite');

const result = parse(navigator.userAgent, collectHints());

Or use parseAsync() to also pick up high-entropy values (platformVersion, architecture, model, full version list):

const { parseAsync } = require('ua-lite');
const result = await parseAsync();

Server-side

Pass the raw Sec-CH-UA-* request headers straight through:

const { parse } = require('ua-lite');

app.get('/', (req, res) => {
  const result = parse(req.headers['user-agent'], req.headers);
  res.json(result);
});

Recognized header keys (case-insensitive):

Header	Effect
sec-ch-ua	Overrides browser.name from the brand list (skips Chromium/Not-A.Brand noise entries)
sec-ch-ua-full-version-list	Overrides browser.name and browser.version with the full version
sec-ch-ua-platform	Overrides os.name
sec-ch-ua-platform-version	Overrides os.version
sec-ch-ua-mobile	?1 sets device.type to mobile, ?0 sets it to desktop
sec-ch-ua-arch	Overrides cpu.architecture
sec-ch-ua-model	Overrides device.model (only when non-empty)

Agent types explained

agent.type classifies what kind of client is making the request — independent of, and detected before, browser parsing:

Type	Covers	Examples
ai	Crawlers that train or answer from AI models	GPTBot, ClaudeBot, Google-Extended, PerplexityBot, CCBot, Bytespider
crawler	Traditional search/indexing/SEO crawlers	Googlebot, Bingbot, AhrefsBot, Twitterbot
bot	Generic scripts and HTTP clients	curl, wget, python-requests, axios, Go-http-client
app	In-app / wrapper browsers	Instagram, Facebook, TikTok, Twitter, WhatsApp, Electron
browser	Everything else — a regular end-user browser	Chrome, Firefox, Safari, Edge

When agent.type is browser, agent.name mirrors browser.name. device.type is forced to 'bot' whenever agent.type is anything other than browser or app.

How to extend

All detection rules live in flat, ordered tables inside src/ua-lite.js — no build step, so edits take effect immediately.

To add a new agent pattern (e.g. a new AI crawler), find the AGENT_RULES array and add a [regex], [['type', 'ai'], ['name', 'YourBotName']], pair in the appropriate tier. Order matters — rules are tried top to bottom and the first match wins, so put more specific patterns before generic ones.

The same pattern applies to BROWSER_RULES, OS_RULES(), CPU_RULES, and DEVICE_RULES_PASS1.

Benchmarked against other popular parsers

ua-lite's domain knowledge (regex ordering, the Windows NT version map, the rgxMapper pattern) was studied from ua-parser-js — see src/ua-lite.js for the disclosure. No code was copied; all patterns here are written fresh. We benchmarked ua-lite against that library (both major versions) plus four other widely used UA-parsing libraries on npm: bowser, platform.js, express-useragent, and mobile-detect.

All figures below are real measurements: each package was installed fresh from npm, required directly (no bundler), and run through the same 12-UA mix (desktop/mobile/tablet/bot) on Node 22.

ua-parser-js v1 and v2 are shown separately because their license terms differ sharply — that difference is the main reason ua-lite exists.

	ua-lite	ua-parser-js v1	ua-parser-js v2
License	MIT	MIT (unmaintained)	AGPL-3.0 (commercial tiers available)
Bot/AI agent detection	✅ built-in, free	❌	✅, but gated behind AGPL copyleft or a paid commercial license
Client Hints	✅ built-in	➖ separate package	✅
Build step	None	None	Requires build
Bundle	Single file	Single file	Multi-file, plus a large bundled icon set

Bundle size (gzip, the actual file require() resolves to):

	ua-lite	bowser	platform.js	express-useragent	mobile-detect	ua-parser-js v1	ua-parser-js v2
gzip size	6.0 KB	7.6 KB	11.9 KB	8.4 KB	22.5 KB	8.7 KB	13.6 KB

Parse throughput (50,000 parses, warm, parses/sec):

	ua-lite	bowser	platform.js	express-useragent	mobile-detect	ua-parser-js v1	ua-parser-js v2
parses/sec	292,000	256,000	22,000	380,000	119,000	88,600	21,500

express-useragent edges out ua-lite on raw throughput — it returns a flatter, simpler object (mostly boolean flags) with less string-shaping per parse. platform.js is by far the slowest of the group, at roughly 1/17th of express-useragent's throughput.

Feature coverage — this is where the field narrows fast:

	ua-lite	bowser	platform.js	express-useragent	mobile-detect
License	MIT	MIT	MIT	MIT	MIT
Browser name/version	✅	✅	✅	✅	❌
Engine (Blink/Gecko/WebKit)	✅	✅	✅ (as "layout")	❌	❌
OS name/version	✅	✅ (empty for bot UAs, correctly — no OS token to find)	✅	✅	✅ (OS family only)
CPU architecture	✅ (arch family: amd64/arm64/arm/...)	❌	➖ (bit-width only — 32/64, misses arch family entirely, even for aarch64)	❌	❌
Device type (mobile/tablet/desktop/tv/console)	✅	➖ (type: "bot" only)	❌	➖ (boolean flags, no unified type)	➖ (mobile/tablet booleans only)
Traditional crawler detection (Googlebot, Bingbot)	✅	✅ (with vendor)	❌	✅	❌
AI-crawler detection (GPTBot, ClaudeBot, PerplexityBot)	✅	✅ (with vendor)	❌	❌	❌
Generic script detection (curl, python-requests)	✅	❌	❌	✅	❌
In-app browser detection (Instagram, Facebook, TikTok)	✅	➖ (device only, no wrapper flag)	❌	➖ (Facebook flag only)	❌
Client Hints (sec-ch-ua-*) parsing	✅	❌	❌	➖ (field exists, must be fed manually)	❌

Takeaways from actually running these side by side:

• bowser is the strongest of the four alternatives on bot detection — it tags many known AI crawlers with a vendor field ("Anthropic", "OpenAI") and returns proper OS name/version for regular browsers — but it has no CPU architecture at all, and no generic-script detection (curl, python-requests come back with an empty browser name, not flagged as bots).
• platform.js is a legacy general-purpose detector: solid browser/OS fields, a CPU field that only reports 32/64-bit (never resolves aarch64 to an actual arm64 family), no bot detection of any kind, and it's ~13–17x slower than the others in this table — it does noticeably more string work per parse than the others.
• express-useragent is the fastest here and has real crawler/script detection (isBot, botName, isCurl), but it predates the AI-crawler wave — it correctly flags Googlebot and curl, but returns nothing for GPTBot or ClaudeBot. It also has a clientHints field in its output shape, but nothing parses sec-ch-ua-* headers into it automatically.
• mobile-detect is a mobile-detection specialist, not a general UA parser — no browser name, no engine, no bot detection at all. It answers one question well ("is this a phone/tablet?") and nothing else.
• ua-parser-js v1 returns full browser/engine/OS/CPU/device fields under a permissive MIT license, but is unmaintained and has no bot or AI-agent classification at all — the gap ua-lite was built to fill.
• ua-parser-js v2 does have AI-crawler pattern lists as rich as ua-lite's, but the whole package (including that submodule) ships under AGPL-3.0. Using it in a closed-source service means either open-sourcing that service or buying a commercial license — a legal cost, not a runtime one. It's also the heaviest and slowest option in this whole comparison.

ua-lite is the only one of the six that combines full field coverage (browser/engine/OS/CPU/device), tiered bot/AI/app classification, and native Client Hints resolution in a single MIT-licensed file with no copyleft strings attached — the closest alternative on features (bowser) still needs OS/CPU data filled in from elsewhere, the fastest alternative (express-useragent) needs AI-crawler patterns added by hand, and the only other option with comparable AI-crawler coverage (ua-parser-js v2) requires either open-sourcing your own product or paying for a license.

License

MIT