Privacy and security
What leaves your machine, and what never does.
Written for the person who has to sign this off, not for the person who installs it. Every call the product makes is below, with what it carries.
The short version
Ad-Preflight checks ad creatives on the machine they are already on. The free check on this site runs inside your browser; the desktop app, its auto-fix, its ZIPs and its Deep Audit run on your own computer. We never receive a creative - not the files, not their names, not the verdicts, not the reports. There is nothing for us to store, nothing for us to leak, and nothing for you to disclose.
The network is used for your licence and for the app's own updates. If you read one thing here, read what uses the network.
The free check on this site
The ZIP, folder or video file you drop is read by this page's own code, in your browser's memory, and is thrown away when you leave. Nothing from it is sent anywhere, and there is nowhere for it to be sent to: this site has no endpoint that accepts a creative or a video.
You do not have to take that on trust. Open your browser's developer tools, switch to the Network tab, and run the check. Drop a creative and watch the list: no request carries the file, and the verdict appears with nothing having gone out.
The one thing the free check does send
A count. One short request says that a check happened and what the verdict was - cleared, hold, grounded, or, on a video, partial - and whether this is the first time this browser has sent that event. That is an event name and a flag: no file, no file name, no size, nothing read out of the creative, and no identifier of any kind. It is sent at most once per browser per day, and skipped entirely when your browser sends Do Not Track or Global Privacy Control.
The same kind of count, under the same rules, is sent when you press a button on this page - the app download, sending a result to us, copying or saving a result or a suggested fix - and when the Windows app is offered to a browser that is not on Windows. That last one says "an Apple device" or "not Windows" and nothing else about the machine: not the browser, not the version, not the screen. A result you copy or save goes to your own clipboard or your own disk, and the count says only that the button was pressed.
One more count goes with a check that ran: which family of system the browser is on - Windows, Mac, iOS, Android or other. It tells us whether the people who try this are people a Windows app can serve, and it is the family name only: no version, no model, no user-agent.
A video is read the same way, and less of it than you would expect
The Video asset tab reads the file's own header - the bytes the container puts at its front to say what it holds. On an MP4 that is all it reads: the box headers and the moov atom, a few hundred kilobytes, through the browser's own file API. A two-gigabyte master is never copied into the page, let alone out of it. The tab also hands the file to a <video> element to see what this browser would play; that decoding happens on your machine, like everything else here.
There is one request, and only if you ask for it: pressing Run the example video fetches this site's own sample file, the same way any image on this page is fetched. It carries nothing about you and nothing about any file of yours.
What that tab cannot do is written into the result rather than left out of it. Loudness, true peak and per-stream bitrates are measurements taken off decoded audio and video, and a browser tab has no decoder to take them with. Those rows come back marked as not measured, with the measurement named - never as passes, and never as blank lines - and a run where everything readable passed but something could not be read is reported as partial, not as clean.
The ad tag mode is different, and says so
A tag cannot be checked without being run. So the tag you paste is executed in an iframe in your own browser, and that calls the ad server the tag names - exactly as a real impression would, and it may well count as one. That request is between your browser and that ad server; nothing in it reaches us, and the text of the tag does not leave your browser either.
If you send a result to be looked at
Under a result there is an offer to answer a rejection in writing. Pressing it turns the findings into plain text in your browser - verdicts, rule names and the file names inside the creative - and puts that text in the form, where you can read every line of it before sending, and remove it. The form accepts no files at all.
What leaves your machine
Nothing about a creative, ever. Specifically, none of this is transmitted anywhere by the free check, the desktop app or the CLI:
- the creative’s files - HTML, CSS, JavaScript, images, fonts
- the video file you check - no part of it, not one frame, not its header
- their file names, folder names, paths or ZIP structure
- the ad tag you paste into the free check or the app
- any verdict, finding, warning or check result
- anything read out of a video - its dimensions, duration, codec, bitrate or audio track
- the screenshots the Deep Audit takes, and the PDF report
- the packaged ZIP, or the backup the auto-fix writes beside your originals
- the landing-page URLs or click macros found inside the creative
There is no "send for analysis" step, no queue, no worker and no bucket, because there is no server-side audit to send anything to. The validator, the auto-fixer, the packager, the preview and the Deep Audit's browser all run as processes on your own hardware.
What uses the network
Five calls, all of them about your licence. None carries a creative or a result.
| What | When | What it sends | What comes back |
|---|---|---|---|
| Activate a licence or a trial | Once per machine | The licence key or the trial’s email address; a hashed machine ID; the OS, CPU architecture and runtime version; and the machine’s own hostname as a label, so you can tell your seats apart | A signed licence file for that machine, and where to fetch the Deep Audit engine |
| Download the Deep Audit engine | Once per machine, after activating | Nothing about you beyond the request itself | The engine - about 300 MB, because it carries a real browser - checked against a published SHA-256 as it arrives |
| Check in | Occasionally, while a paid licence is in use | The licence key and the hashed machine ID. Nothing else | Whether the licence is still active, and a refreshed licence file |
| Release a seat | When you ask, to move the licence to another machine | The licence key and the hashed machine ID | Confirmation that the seat is free |
| Recover a key | When you ask | The email address you type | Nothing to the caller: the key is emailed to that address if it has one. The answer is identical either way, so the call cannot be used to find out who our customers are |
The hashed machine ID is a SHA-256 of the machine's own identifier, which is what ties a seat to a machine so a per-machine licence can be enforced at all. It is not reversible to a serial number, a user, a network address or a person.
The desktop app's two extra calls
- An update check, on start and when you press Check for updates. With usage counts switched on it adds two headers: the app's version, and whether this start is a first run, an update or a reopen. Counts only, with no identifier.
- Anonymous usage counts, at most once a day per event while they are switched on: which offers were seen or pressed, that an audit ran, and two coarse ranges about how long the app has been in use. No machine ID, no install ID, no email, nothing about your files or your results. Switch them off in Settings, under "Help us do better", and one last note says they were switched off - then nothing at all.
Ranges rather than numbers, and once a day rather than per action, are deliberate: what comes out is a count of apps that did something, with no way to follow one machine through it. The CLI has neither - no usage counts, no update check, nothing of its own.
One call that is not ours
A creative that loads a font or a pixel from a CDN still loads it: during a Deep Audit the local browser makes that request because the creative asks it to, just as the publisher's page would. That is a request from your machine to the host the creative names, and we are not in it - reporting those requests back to you is one of the Deep Audit's checks. The same is true of the ad tag mode above.
What needs a connection
Ad-Preflight is not an offline tool, and this page is not going to claim it is. The guarantee is about what is transmitted, not about running with the network unplugged.
| What | Needs a connection |
|---|---|
| The free check on this site, on a creative ZIP or folder | Only to load this page. The check itself requests nothing |
| The free check on this site, on a video file | Only to load this page. The check itself requests nothing; pressing “Run the example video” fetches this site’s own sample file, and nothing else |
| The free check on an ad tag | Yes, by definition - the tag is loaded from the ad server it names |
| The app’s standard audit of a creative | No |
| Activating a licence or a trial, recovering a key, releasing a seat | Yes - the calls listed above |
| A trial Deep Audit | Yes, every time it runs |
| A paid Deep Audit | Not per run. The machine does have to reconnect periodically to stay licensed |
| The desktop app’s update check | Yes, on start |
| A creative that loads a remote font, pixel or asset | Yes, during a Deep Audit - your own browser fetches it from that host, as the publisher’s page would |
None of those sends us your creative, your verdicts or your report. That part holds whether you are online or not.
Why there is nothing here to put in an NDA
An NDA covers disclosure. Using Ad-Preflight involves no disclosure of the creative to us: the processing happens on your own equipment, under your own control, and we are not a recipient, a processor or a sub-processor of the creative material at any point.
That is a structural statement, not a policy we could quietly change. There is no endpoint for a creative to be sent to. A promise not to look at uploaded files depends on conduct; having no upload depends on architecture.
- Creative material: never transmitted. No controller or processor relationship arises over it.
- Licence data: we are the controller for the small set of records below - the purchase email, the licence key, hashed machine IDs and machine labels - which exist to enforce a per-machine licence and to let you recover your key.
- Questionnaires: send them. The answers are short because the architecture is.
What is kept
- Creatives: none. We never receive them, so there is nothing to retain, nothing to delete on request and nothing to produce under a subpoena.
- Licence records are kept while the licence exists, because a per-machine licence cannot be enforced without them: the purchase email, the licence key, hashed machine IDs and the machine labels you see in Settings. Releasing a seat removes that machine's record.
- Counts - from the app and from the free check alike - are stored as totals per day, with no identifier in them to resolve back to a machine, an install or a person.
- This website's own statistics are counted on the server as daily totals: which page was viewed, which kind of site a visit came from (a search engine, a link somewhere, or none), and which family of system the browser said it runs on (Windows, Mac, iOS, Android or other). The full referrer, the user-agent string and the query string are not kept, no cookie is set for them, and nobody's path through the site can be rebuilt from them.
- The pages also load Cloudflare Web Analytics, which is how we see how fast they are. It is a cookieless script: it reports the page address, the referrer and the load timings to Cloudflare, which already serves this site. It sets no cookie, follows nobody from site to site, and carries nothing from the check - it runs on every page, whether or not you use one.
- A rejection you send us - the text you typed and the findings your browser produced - is used only to answer you, is never shared or published, and is deleted within 30 days of the reply.
The privacy policy is the legal version of the same facts, and the terms cover the licence itself.
Questions
- Do my creatives get uploaded?
- No. The free check on this site reads the ZIP, the folder or the video file inside your browser and sends nothing from any of them; the desktop app reads the folder you point it at and writes the verdict, the repaired files and the ZIP back beside it. The only things that use the network are about your licence and the app itself, never about a creative.
- Do I need an account?
- No. The free check has no sign-up and no key. The desktop app installs and runs. Buying Pro gives you a licence key you activate on a machine; there is still no account to log into and no project to join.
- Do we need an NDA or a DPA before I use this on a client’s campaign?
- For the audit there is nothing to cover: we never receive the creative, so there is no disclosure to us and no processing by us. We are the controller only for the licence records - the purchase email, the licence key, hashed machine IDs and the machine labels you see in Settings. If your legal team wants that in writing, send us the questionnaire and we answer it.
- Do I need to be online?
- Usually yes, and we would rather say so than have you find out. Activating a licence needs a connection, a trial Deep Audit needs one every run, the desktop app checks for updates on start, and checking an ad tag calls the ad server by definition. What never happens is your creative being transmitted.
- Is there a limit on how many creatives I can check?
- No. Nothing is metered per asset or per run, and nothing phones home to count one. A licence covers a machine, not a quota, so a forty-size campaign costs what one banner costs, however many times you re-run it.
- Can our security team review the code?
- The repository is private, so there is no public source to review. What your team can do is pin the published npm package in a lockfile or mirror it in an internal registry, so the version that ran yesterday is the version that runs today. For the question that matters about a creative - where it goes - the answer is watchable on this page: open your browser’s Network tab and run the free check on a creative ZIP.
- What happens if I send you a rejection to look at?
- What arrives is the free check’s result turned into text in your own browser - verdicts, rule names and file names - plus whatever you type. Not the files, and not their code; the form accepts no attachments. We answer by email, and both the email and the record are deleted within 30 days of the reply. If a hard case needs the actual file, we ask in the reply, and you can say no.
See it for yourself
Open the Network tab, drop a creative into the free check, and watch what goes out. Then read the verdict: it is the same audit the desktop app runs, on the same rules.
Run the free check