cyberchef.tools — Privacy Policy
Last updated: 22 September 2026
This policy covers the website at cyberchef.tools — the pages
served from this domain, the app itself and the landing pages beside it at
/extension/ and /relay/. It describes what this
deployment does and nothing else. The browser extension has
its own separate policy.
1. What this site is
This is a static mirror of GCHQ's CyberChef (Crown copyright, released under the Apache-2.0 licence). It is the same application, built from the same source, served as plain files.
It is operated independently by the developer of the CyberChef browser extension. It is not affiliated with, endorsed by, or operated by GCHQ. GCHQ's own copy of CyberChef lives at gchq.github.io/CyberChef, and that is the site to use if you would rather not rely on this one.
The version deployed here is kept in step with the version of the CyberChef engine bundled inside the extension, so that handing a recipe from the extension to this site lands on a matching build.
Three things are added on top of the upstream project: the deployment itself, the analytics described in section 6, and the workbench — a layer of investigation features built on the same engine (recipe shelves and guided walk-throughs, output viewers, log and Sigma tools, a command palette and the rest). The workbench runs in your browser like everything else. Sections 2 to 5 set out exactly what it keeps on your machine, and the one feature that can contact anybody else — only after you switch it on and press the button yourself. No operation and no part of the workbench sends your input, your output or your recipe to this site.
2. Your data stays in your browser
CyberChef performs every operation on your own machine. The whole application is downloaded to your browser and runs there. When you paste text, drop a file in, build a recipe and bake it, the input, the recipe and the output are handled entirely by JavaScript running in your browser tab.
Nothing you paste, type, drop or decode is sent to this server for processing. There is no workbench processing backend. The optional usage analytics in section 6 receives only its documented event fields, never the material you are working on. Optional network operations and Endpoint lookups are described separately below.
What the workbench keeps on your machine
The workbench remembers your settings and your own material so they are
still there when you come back. The IOC lists you import are written
to this browser's IndexedDB, to a database named
cyberchef-workbench belonging to this origin — they are the only
thing kept there. Everything else — saved recipes, recent recipes, your
theme, per-recipe notes and parameters, the workbench's own option toggles
and any Endpoint API keys you enter (section 3) — is written to this
browser's localStorage on this origin.
None of it is uploaded. There is no account to sync it to and no workbench synchronisation service. It is storage on your machine, read only by this page in this browser, and it goes away when you clear this site's data.
One optional feature reads outside that storage: “Offer to peel what I just copied” in Options, which is off by default. With it on, the page reads the clipboard at three deliberate moments — when you return to the tab, when the tab becomes visible, and when you press Cmd/Ctrl+V outside a text box — and your browser asks you for clipboard permission the first time. What it reads is offered to you on a strip under the input and nothing more is done with it: it is not saved, not written to IndexedDB or localStorage, and not sent anywhere. Only a short numeric fingerprint is held, in memory, so the same clipboard is not offered to you twice. Leave the option off and the clipboard is never read at all.
Anything the workbench lets you export — a CSV of scan results, a masked card report, a log export — is built inside the page and handed straight to your browser's download machinery; it is not posted anywhere on the way. Importing a file, such as an IOC list, reads the file you pick, in the page.
Recipes and inputs in the URL
CyberChef can carry a recipe and an input in the address bar, in the part of
the URL after the #:
https://cyberchef.tools/#recipe=From_Base64()&input=<your text, base64>
That part of a URL is called the fragment. Browsers do not include the fragment in the HTTP request they send to a server. The request that reaches this site — and the host in front of it — carries the path and query string, not the fragment. The page reads the fragment locally, in your browser, and processes it there. So a recipe and an input in the address bar are not transmitted to this site in the request.
That is a property of how browsers work, and it does not make such a URL private. The whole URL, fragment included, goes into your browser history — and so into anything that reads your history: profile sync across your devices, session restore, the address bar's suggestions, screen sharing, and anyone with access to your machine. A URL you paste into a chat or an issue tracker carries the input with it, in full, to whoever reads it. Treat a CyberChef link as if it contains the data it contains, because it does. One more consequence of that is set out under page views in section 6.
3. The Endpoint tier — the one feature that can leave your browser
Everything in section 2 is unconditional. The Endpoint tier is the single exception, and it is off by default.
It has to be armed twice before anything is sent. First you switch on “Allow Endpoint lookups with my own API keys” in Options. Then, next to a particular indicator, you press a particular button carrying the amber Endpoint label. Only then, and only for that one indicator, does a request leave your browser. Nothing is looked up in the background, on a timer, or in bulk.
What is sent is the value of the indicator you pressed the button beside — an IP address, a domain, a URL, an MD5, SHA-1 or SHA-256 hash, or a CVE id — carried in the request URL, together with the API key for that vendor in a request header where a key is used. Your input, your output, your recipe and your notes are not part of the request.
Three vendors are contacted directly by your browser:
-
RDAP at
rdap.org— registration data for an IP address or a domain. No key. -
DNS over HTTPS at
cloudflare-dns.com— A, AAAA, MX, TXT and NS records for a domain. No key. -
NVD at
services.nvd.nist.gov— the published details of a CVE id. A key is optional.
Four more cannot be called from a browser at all, because they send no CORS headers. They are reachable only through a relay whose address you configure yourself; with no relay set, their buttons are disabled and say why. When you have set one, the request goes to your relay, and your relay goes to the vendor:
- VirusTotal at
www.virustotal.com— key required. - AbuseIPDB at
api.abuseipdb.com— key required. - urlscan.io at
urlscan.io— key required. - GreyNoise at
api.greynoise.io— key required.
API keys you enter are kept in this browser's localStorage, in clear text, on your own machine, alongside your other preferences. They are not encrypted, they are never sent to this site, and each key is sent only to the vendor it belongs to — or, for the four relayed vendors, to the relay you pointed at that vendor. Anyone who can read this browser profile can read those keys. If that is not acceptable on the machine you are using, do not enter them. The same goes for the optional relay token, a shared secret some relays require: it is kept in localStorage beside the relay address and is sent, as a request header, only to the relay you configured — never to a vendor, and never to this site.
Once a request reaches a vendor it is that vendor's, and the vendor's own privacy policy governs it — not this one. They see your IP address, the value you looked up, and, where a key is used, which account looked it up. If you configured a relay, whoever runs that relay (most likely you) sees the same. This site does not operate a relay; the Worker source and instructions at /relay/ are published for you to deploy yourself. It receives nothing back from these lookups, and keeps no record that you made one: the result is shown to you, on your machine.
4. Working offline, and installing the app
This site registers a service worker — a small script your browser runs on this origin so the app still opens when the network is slow or absent. It caches this site's own files and nothing else: the page, the scripts, the styles, the images. Every request that is not same-origin is ignored by it outright, so it never fetches, caches or observes another site's resources, and it sends nothing anywhere itself. Its cache lives in your browser and is cleared with this site's data.
A few parts of the app are large enough that they are only downloaded when you first use them — the YARA engine behind the YARA scan view is one of them. Like everything else the app loads, it is one of this site's own files, served from this origin and cached by that same service worker; nothing you scan with it leaves your browser.
Because of that service worker and a web app manifest, your browser may offer to install this site as an app. Installing changes where the window lives, not what the code does: the same files run in the same browser sandbox, under everything described in this policy. Installing it and then taking the network away is the simplest way to see for yourself that the app does not need to talk to anybody.
No downloadable copy of this app is offered. There is no zip, no installer and no other package of it to obtain, and nothing on this site links to one; the installable app described above is this site, cached by your own browser. NETWORK.md sets out exactly what this app sends and how to check it from your browser's developer tools, an intercepting proxy or a machine with no network at all.
5. Sharing text to the app, and opening files with it
An installed copy can register with your operating system as a
share target and as a handler for file types
such as .txt, .log, .csv,
.json, .eml and .pem.
Opening a file with CyberChef keeps that file on your machine. Your browser hands the page a handle to the file you chose, the page reads it locally and puts its contents into an input tab. The contents are never transmitted, to this site or to anyone else.
Sharing text to CyberChef also keeps that text on your machine.
The share target is a POST, and the request it makes never leaves
your browser: it is answered by the service worker from section 4, which is
this site's own code running locally on this origin. The service worker reads
the shared text out of the form body, hands it back to the page in the URL
fragment as #input=<your text, base64>, and the page loads
it into an input tab. The shared text is never sent to the host,
so it does not appear in this site's hosting request logs at Cloudflare
(section 7), and it is not included in any analytics event, because the
analytics excludes shared text and URL fragments (section 6).
What it does become is part of a CyberChef link, with everything section 2 says about that: the fragment goes into your browser history and into the address bar, like any URL carrying an input. If the text you were about to share is sensitive, that is the thing to weigh — or paste it into the input instead of sharing it.
6. Analytics
The hosted site offers optional first-party product analytics.
After you select Allow analytics, fixed usage events go to
/api/analytics on this domain. A Cloudflare Pages Function
validates them and stores them in Cloudflare D1. A private dashboard,
protected by Cloudflare Access, displays reports to the site operator.
Cloudflare hosts and processes this information; first-party does not mean
that analytics stays on your device.
Before you allow analytics, no analytics visitor identifier is created and no usage events are sent. The Analytics settings control lets you change your choice. Global Privacy Control and Do Not Track disable collection. The workbench continues to work with analytics declined, blocked, or unavailable. Offline builds do not load this analytics runtime.
Existing Google Analytics reports remain available separately to the operator. Their historical aggregate archives in D1 have no automatic expiry. First-party and Google counts are never added together, and the first-party visitor and session identifiers are never shared with Google.
What is sent
-
Page views and engagement.
page_viewhas no event parameters. Its request context records a fixed, authored page path; an unrecognised path becomes a fixed fallback. No query string, URL fragment, raw referrer, page title or user-supplied URL is transmitted.engagementmeasures time spent with the page visible. Its one parameter,duration, is a fixed bucket:under_10s,10_to_30s,30_to_60sorover_60s. It does not record interactions or text. Acquisition is a coarse source category, not a referring URL. -
Twenty-five product event types, and only these twenty-five.
Their parameters are the fixed labels and counts described below:
-
input_activity— sent at most once per page load, when the input first becomes non-empty. It has no parameters and does not include the input, its size or how it arrived. -
bake— sent when you click the Bake button. Its one parameter,operations, is a number: how many operations are in the recipe. Not the operations, not the arguments, not the input. Just the count. -
operation_add— sent when you add an operation to the recipe. Its one parameter,operation, is the operation's name as it appears in the list, for exampleFrom Base64. It is sent when the operation is added, before you have typed anything into it, and it never carries the operation's arguments. -
recipe_save— sent when you save a recipe to your browser's local storage. It has no parameters at all. The recipe itself is saved in your browser and is not sent. -
recipe_load— sent when you load a saved or pasted recipe. Its one parameter,operations, is again a number: how many operations that recipe contains. -
workbench_section_open— sent when the Recipes, Learn or Tools section opens. Itssectionparameter is one of those three fixed labels. It does not include a search or anything in the section. -
workbench_recipe_use— sent when a recipe from the workbench library is applied. Itsoperationsparameter is only the number of operations in the resulting recipe. It does not include the recipe's name, source, operations, arguments or input. -
workbench_learn— sent when a local challenge is solved. Itsactionparameter is the fixed labelsolve. It does not include the challenge or answer. -
workbench_tool_open— sent when a workbench tool opens. Itstoolparameter is a fixed internal id from the Tools grid, such aslogsorregex. It does not include the tool's input, settings, query or results. -
tutorial_beginandtutorial_complete— the fixed events sent when any local walkthrough starts and finishes. They have no parameters, so the walkthrough and its steps are not identified. -
magic_mode— sent when automatic Magic is switched on or off. Itsstateparameter is the fixed labelenabledordisabled. -
magic_result— sent when a Magic attempt reaches one of the outcomes below. Superseded or cancelled searches are not sent. Itsoutcomeparameter combines the fixed invocation labelmanualorautomaticwith one ofsuccess,no_input,too_large,no_match,timeoutorerror. It does not include the input, output, discovered recipe, error text or any candidate Magic considered. -
output_action— sent when the fixed Copy output, Download output or Use output as input control is selected. Itsactionparameter iscopy,downloadoruse_as_input. Output contents and filenames are not sent. -
suggestion_use— sent when an input suggestion is selected. Itskindparameter is onlyrecipe,toolormagic. The suggestion, input, detection evidence and selected recipe or tool are not sent. -
workbench_tool_complete— sent at most once after each time a tool is opened and completes its primary action successfully. Itstoolparameter uses the same fixed Tools grid id asworkbench_tool_open. Inputs, settings and results are not sent. -
web_vital— sent at most once for each supported Core Web Vital when the page is hidden. It contains the fixed metric labellcp,clsorinp, Google's fixedgood,needs_improvementorpoorrating, and the same rating as a fixed metric bucket. Raw timings, DOM content and interaction details are not sent. -
client_error— sent at most once for each coarse source:runtime,resourceorpromise. Error messages, stack traces, filenames and URLs are not sent. -
first_useful_output— sent once after the first rendered bake following non-empty input. Itslatencyparameter is onlyunder_1s,1_to_3s,3_to_10sorover_10s. -
bake_failure— sent at most once for each fixed failure category,operationorworker. Error text, operation arguments, recipes, inputs and outputs are not sent. -
surface_weborsurface_extension— one fixed event labels whether this load is the normal website or the website framed by the browser extension's thin client. Both are part of the same CyberChef product totals. The fixedreleaseparameter contains the public application version. -
traffic_production,traffic_previewortraffic_internal— one fixed event labels a production load, apages.devpreview, or a browser that the operator explicitly marked as internal in local storage. It has the same fixedreleaseversion parameter.
-
- Random browser, session and page identifiers. After consent, a random visitor identifier is stored in localStorage for at most 365 days. It recognises repeat visits from this browser, not your name or account, and does not connect your different devices. Sessions expire after 30 minutes of inactivity or 24 hours total; each page load and session renewal has a fresh page identifier. Events have random identifiers to prevent duplicate counting on retries, a page-local sequence and a bounded elapsed time. The collector assigns its own receipt time. Clearing browser storage creates a new visitor on a later consented visit.
-
Browser and device context. Fixed browser and operating
system families, browser major version, device class, screen and viewport
size buckets, language and timezone help us find compatibility and
performance problems. Country, where available, comes from Cloudflare's
request metadata. The analytics database does not store IP addresses,
full user-agent strings, raw referrers, precise location or fingerprints.
Context fields are
surface,trafficClass,release,pagePath,pageCategory,referrerCategory,browserFamily,browserMajor,osFamily,deviceClass,viewportBucket,screenBucket,languageandtimezone. Paths, categories and browser/device labels come from fixed lists; language and timezone are validated against supported values. No arbitrary properties are accepted. A fixed surface identifies the website or hosted extension; a traffic label separates production, preview and operator testing. - Network metadata. Cloudflare necessarily receives your IP address and request headers when your browser contacts this site, including its analytics endpoint. This hosting traffic is separate from the fields stored in the product analytics database. See section 7.
What is never sent
Your input. Your output. The arguments you type into an
operation. The contents of a file you load. The text of a recipe.
None of it appears in any analytics event: the events above are the complete
list, and their parameters are counts or fixed labels, apart from the
existing operation name sent by operation_add. There is no code
on this site that reads your input or output for analytics purposes.
Searches, recipe names and contents, walkthrough and challenge identities, files and filenames, viewer activity, command-palette text, Endpoint activity, Magic candidates, Magic output and tool results are not sent. Magic telemetry contains only the fixed state and outcome labels listed above.
Retention and your choice
First-party event and session records are retained for 90 days, visitor first/last-seen metadata for at most 365 days, and daily aggregate history for 730 days. Hashed visitor identifiers used for rate limiting expire after two days. A daily scheduled job removes expired records. Reports count browser identifiers; they cannot identify unique people and do not include visitors who decline or block analytics.
- Decline or withdraw through Analytics settings. Withdrawal stops collection, clears the local analytics identifiers and discards unsent events. A preference is kept locally to remember your choice. Withdrawal does not erase records already received; those expire under the retention periods above.
-
Browser privacy controls and blockers work. Global Privacy
Control or Do Not Track disables this site's collection. A content blocker
can also block
/api/analytics. First-party analytics is not unblockable and the workbench never requires it. - The private offline workbench has no analytics. The public extension frames the hosted site and uses the same analytics choice; its private offline workbench runs from bundled files without this runtime.
- Upstream CyberChef is available separately. GCHQ's standalone copy runs locally under its own build configuration.
7. Hosting
This site is served by Cloudflare Pages. Like any web host, Cloudflare sees the requests your browser makes in order to answer them, and records request metadata in its logs — your IP address, your user agent, and the path requested. That processing is Cloudflare's, under Cloudflare's privacy policy.
Cloudflare also serves some of its own platform code from paths beginning
/cdn-cgi/ on this domain — its bot-detection script, for
example — and may set its own cookies for security purposes. That is the
host's platform rather than the CyberChef application, and it is covered by
the same Cloudflare policy.
The application itself remains a client-side workbench. Cloudflare Pages Functions receives optional analytics events, and a protected administrator route reads first-party reports and historical Google Analytics reports. D1 stores analytics as described in section 6. These routes do not receive or store CyberChef inputs, outputs, recipes or files.
8. Cookies and browser storage
First-party analytics uses localStorage for its consent preference and, only after consent, random visitor/session identifiers. These identifiers are sent only in the documented analytics request body. They are not authentication credentials. See section 6 for expiry and withdrawal.
Cookies from earlier Google Analytics visits may remain until they expire or you clear this site's data. Cloudflare may separately set security and bot-detection cookies as the host; see section 7.
Saved recipes, themes, interface preferences and Endpoint keys remain local workbench data, and imported IOC lists remain in IndexedDB, as described in section 2. Analytics never reads that material. Clearing site data erases local workbench storage and the analytics preference and identifiers; it does not delete analytics records already received by the server.
9. The browser extension
There is a CyberChef browser extension by the same developer. Its public
version is a thin client: extension controls collect the selected page text
locally and frame cyberchef.tools for the workbench. Operations
still run in that framed page in the browser. Because it is the same hosted
site, the analytics choice in section 6 applies. A consented load is counted in the
total product figures and under the fixed surface_extension
label. The separately distributed private offline workbench uses bundled
files and does not load the hosted analytics script.
- About the extension
- The extension's privacy policy — a separate document covering separate software
This site is one of the endpoints the extension can be pointed at, and it is never the default. What the extension sends here, and only when you ask it to, is described in its own policy.
10. Changes to this policy
If what this site does with data changes, this policy changes with it, and the "last updated" date at the top changes too. The current version is always the one published at https://cyberchef.tools/privacy.
11. Contact
Questions, corrections, or something here that does not match what you observe: [CONTACT].
(A public contact address for this site has not been chosen yet. This placeholder will be replaced with one.)