Free tool
What is my user agent?
This is the user agent string your browser sent, the other headers sites use to tell people from AI agents, and how Rebilder's open-source classifier reads them. Paste any user agent below to check it too.
Your user agent
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)
How this request is classified
- Kind
- AI assistant
- Software working for a person or an AI company: an assistant reading a page for someone, a coding agent, or an AI crawler. A site running the Rebilder gateway can give it a clean text version of the page.
- Platform
- Claude
- The company or product the request says it comes from.
- Confidence
- Medium
- The user agent decided it. That is a claim the client makes about itself, and any client can copy it.
Signals
ua:claudebotThe user agent containsclaudebot, a known agent token for Claude.
Headers that matter for agent detection
| Header | Sent with this request | What it tells a server |
|---|---|---|
| Accept | */* | The formats the client will take. Browsers ask for HTML; agents that want a clean text version list text/markdown. |
| Accept-Language | Not sent | Your language preferences. Browsers almost always send it; many crawlers and scripts do not. |
| Sec-CH-UA | Not sent | Chromium browsers name their brand and major version here. |
| Sec-CH-UA-Mobile | Not sent | ?1 on a phone or tablet, ?0 elsewhere. |
| Sec-CH-UA-Platform | Not sent | The operating system, such as "Windows" or "Android". |
| Signature-Agent | Not sent | Names the operator of a signed agent, such as "https://chatgpt.com". Browsers never send it. |
| Signature | Not sent | An HTTP message signature (RFC 9421). Used by Web Bot Auth. |
| Signature-Input | Not sent | What the signature covers, its key and when it expires. |
The checker shows these headers only to you. It does not log or store them, and a user agent you paste below is checked in your browser and never sent to us.
Check any user agent
Paste a user agent from your server logs, or pick an example, to see whether it reads as a person, an AI agent or a crawler.
- Kind
- AI assistant
- Software working for a person or an AI company: an assistant reading a page for someone, a coding agent, or an AI crawler. A site running the Rebilder gateway can give it a clean text version of the page.
- Platform
- ChatGPT
- The company or product the request says it comes from.
- Confidence
- Medium
- The user agent decided it. That is a claim the client makes about itself, and any client can copy it.
Signals
ua:gptbotThe user agent containsgptbot, a known agent token for ChatGPT.
What is a user agent?
A user agent is the software that makes a web request on your behalf, and the user agent string is how it introduces itself. Every request carries it in the User-Agent header, so the server can tell a phone from a laptop, a browser from a script, and a person from a crawler.
A current Chrome user agent reads like this: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/142.0.0.0 Safari/537.36. Most of it is history:
| Part | What it means |
|---|---|
Mozilla/5.0 | A compatibility token every modern browser sends, left over from the 1990s. It no longer means Firefox. |
(Windows NT 10.0; Win64; x64) | The operating system and CPU. |
AppleWebKit/537.36 (KHTML, like Gecko) | The rendering engine family, again kept for compatibility. |
Chrome/142.0.0.0 | The browser and its major version. Chrome now reports zeros for the minor parts. |
Safari/537.36 | One more compatibility token. |
Chromium browsers now send a reduced user agent and move the detail into client hints: Sec-CH-UA names the brand and version, Sec-CH-UA-Mobile says whether it is a phone, and Sec-CH-UA-Platform names the operating system. Safari and Firefox do not send them, which is why the table above may show them as not sent.
How AI agents identify themselves
AI crawlers and assistants add a product token to their user agent, usually after the word compatible. The token is what you match in logs, in robots.txt and in firewall rules. Each operator uses separate tokens for separate jobs:
| Token | Operator | What it does |
|---|---|---|
GPTBot | OpenAI | Collects pages that may be used to train OpenAI’s models. |
OAI-SearchBot | OpenAI | Indexes pages for ChatGPT search. See OAI-SearchBot. |
ChatGPT-User | OpenAI | Fetches a page while a ChatGPT user asks about it. |
ClaudeBot | Anthropic | Collects pages for training. See ClaudeBot. |
Claude-User, Claude-SearchBot | Anthropic | Fetches a page for a Claude user, and indexes pages for Claude’s search. |
PerplexityBot, Perplexity-User | Perplexity | Search indexing and user fetches. |
meta-externalagent, meta-webindexer | Meta | Training and product indexing, and Meta AI search. |
Bytespider | ByteDance | An AI crawler, per Cloudflare’s bot reference. |
Googlebot, bingbot, Applebot | Google, Microsoft, Apple | Search crawlers. Rebilder classifies them as crawlers, never as agents. |
Coding agents identify themselves differently. Claude Code sends claude-code/2.0.13 (external, cli) and asks for text/markdown in its Accept header. The AI agent user agents reference has the full strings and how to match each one.
Why a user agent is a claim, not proof
Any client can send any user agent. A script can call itself Googlebot, and a crawler can call itself Chrome. Paste Mozilla/5.0 (compatible; GPTBot/1.2) above and it classifies as GPTBot, because the string is all the parser has. That is why Rebilder caps a user-agent match at medium confidence.
- An
Accept: text/markdownheader states what the client wants. It is a preference, not an identity, but it is the clearest sign that an agent is asking. - Web Bot Auth lets an agent sign each request (
Signature,Signature-Input) and name its operator (Signature-Agent). A server that checks the signature against the operator’s published key knows who sent it. - Published IP ranges confirm the large crawlers: Google, Microsoft, OpenAI and Anthropic publish the addresses their bots use, and Google and Microsoft also support a reverse DNS check.
Use the user agent to choose a format for the same content. Anything that depends on who the caller really is, such as private prices or account actions, needs a verified signature or a login. Classification documents the full order of checks.
How to read the result
The checker runs detect() from @rebilder/agent-detect, the open-source classifier inside the Rebilder gateway. It reports four things:
- Kind: a person, an AI assistant or agent, a crawler, or an agent protocol client. Known crawlers always classify as crawlers, even when they ask for markdown.
- Platform: the company or product the request names, such as ChatGPT, Claude or Googlebot.
- Confidence: high for an explicit header signal or an ordinary browser, medium for a user-agent match, low when there was little to go on.
- Signals: every check that fired, in order, so you can see why it reached its answer.
The pasted-string parser sees only the user agent. The result for your own request also reads your Accept and signature headers, so the two can differ.
Decide what AI agents can read
Once you know the tokens, you can choose which crawlers may read your site. The AI robots.txt generator writes the rules by purpose: decline model training, stay in AI search, and let assistants fetch a page when a customer asks.
For the agents you allow, the Rebilder gateway answers at the same URL with clean markdown built from your own product, service and policy data, while people and search crawlers keep your HTML.
Frequently asked questions
What is my user agent?
It is the User-Agent header your browser sent with this request, shown at the top of this page. It names your browser, its version and your operating system, and every site you visit receives it.
How do I find my user agent in my browser?
Open this page. To read it yourself, open the developer tools console and run navigator.userAgent, or look at the request headers in the Network panel.
What is a user agent parser?
A tool that reads a user agent string and says what sent it. This one reports whether the string belongs to a person’s browser, an AI agent or a crawler, and which platform, using the same classifier the Rebilder gateway runs.
Can a user agent be faked?
Yes. Any client can send any string, so a user agent is a claim. Large crawlers can be confirmed by reverse DNS or published IP ranges, and agents that sign requests with Web Bot Auth can be verified against their operator’s key.
What user agent do AI crawlers use?
Each operator adds its own token: GPTBot, OAI-SearchBot and ChatGPT-User for OpenAI; ClaudeBot and Claude-User for Anthropic; PerplexityBot for Perplexity; meta-externalagent for Meta; Bytespider for ByteDance. The reference lists the full strings.