What a user agent parser reads
A user agent is a line of text that a browser or program sends with every web request. A user agent parser splits that line into parts: the browser and its version, the rendering engine, the operating system, the kind of device and sometimes the model. Take the Example string:
Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36 Edg/124.0.0.0Read from the right, Edg/124.0.0.0 says it is Microsoft Edge 124. Chrome/124.0.0.0 and AppleWebKit are there because Edge is built on the same engine as Chrome, which is Blink. Windows NT 10.0 and Win64; x64 give the system and the 64-bit processor. Mozilla/5.0 and Safari/537.36 are old labels that nearly every browser still sends so that old websites do not turn them away. This is why a parser has to check the tokens in the right order. A string with Chrome in it is not always Chrome.
Why Chrome shows an old OS version, or a model called K
The user agent string used to carry exact details, and sites used those details to follow people around the web. Chromium-based browsers now keep parts of it fixed. The minor version numbers are zeros, as in 124.0.0.0, and on Android the string says Android 10 and the model K whatever the real phone is. Safari and Chrome on a Mac report macOS 10.15.7 on every newer version, and Windows 10 and Windows 11 both send NT 10.0.
The parser does not guess around this. When it sees these placeholders it says so in the notes, so you know a value like Android 10 or macOS 10.15.7 is not the real version. The real values are available to a website only through Client Hints, which a site has to ask for. The My browser button shows what your own browser reports that way.
Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Mobile Safari/537.36
→ Chrome 124.0.0.0 · Blink · Android 10 · mobile · model KParse many user agents at once from a log
Web servers write the user agent at the end of each access log line, in double quotes. With Input on Detect or on Access log, the tool takes that last quoted field and ignores the rest of the line. A line like this one gives a Chrome on Windows result:
203.0.113.7 - - [07/Nov/2026:10:00:01 +0000] "GET / HTTP/1.1" 200 5123 "-" "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/124.0.0.0 Safari/537.36"Lines whose user agent is a dash are skipped and counted. Input can hold up to 200,000 lines, and lines after that are skipped with a note. The summary lists the most common browsers, systems and device types with their counts and shares, and the table lists each parsed string. The CSV has one column per field, plus the original string, so it opens in a spreadsheet or loads into a database.
How bots and scripts are detected
The parser knows the names of common search, social, SEO, AI and monitoring crawlers, such as Googlebot, Bingbot, GPTBot, ClaudeBot and AhrefsBot, and of HTTP clients such as curl, python-requests, Go-http-client and OkHttp. Any other string that contains a word like bot, crawler or spider is marked as an unknown bot. Headless Chrome is flagged as automation but still shows its browser and system.
Treat the result as a label and not as proof. Anyone can send a string that says Googlebot. To check that a visitor really is Google's crawler, look up the reverse DNS name of its IP address, as Google's documentation describes. The parser says this in the notes for every crawler it finds.
Device type, vendor and model
- Mobile: iPhone and iPod, Android strings with the Mobile token, Windows Phone, KaiOS and Opera Mini.
- Tablet: iPad and Android strings without the Mobile token, Kindle and Silk.
- TV, console and wearable: smart TVs and Fire TV, PlayStation, Xbox and Nintendo, and watches.
- Desktop: Windows, macOS, Linux and Chrome OS strings that do not match any of the above.
- Vendor and model: Apple devices are named from the string. For Android the token before Build or the closing bracket is the model, and the vendor is filled in when the model name matches a known pattern, such as SM- for Samsung and Pixel for Google. Unknown models are shown as they are, with no vendor.
An iPad can send a Mac string when it is in desktop mode, so it is reported as a Mac. A Mac with Apple silicon still says Intel. The notes mention both cases.
Limits of this parser
The detection rules are built into this page and are written for current browsers, the common crawlers and the usual operating systems. A new or rare browser may be reported with its engine only, or as Unknown. Brave and some other Chromium browsers send the same string as Chrome, so they are reported as Chrome. The parser reads the string only. It does not look up the IP address, and it cannot tell that a string was edited.
Everything runs in your browser. The strings and logs are not sent to a server and are not stored. See the Privacy Policy for details.
Related tools
To take apart the URL in a log line, use URL Parser. If your logs are JSON Lines, JSON Lines Viewer opens them, and URL Decode reads percent-encoded paths. To work with the IP addresses in the same logs, CIDR to IP Range and IP Range to CIDR convert between network notations.