The claim is easy to make and hard to verify, so here is the specific list rather than the sentence.
Never stored
Who owns your data is the architecture this follows from, the zero-logging commitment covers what passes through without being kept, and revoking us is how you end it.
| Yours | Where it lives |
|---|---|
| Your users, and their devices | Your D1, pithy_auth_* |
| Their purchases and subscriptions | Your D1, pithy_payments_* |
| Their entitlements | Your D1 |
| Your audit events | Your D1, pithy_audit_* |
| Your support mail, and its attachments | Your D1 and your R2 |
| Your email jobs and suppression list | Your D1 |
| Your secrets, at any point | Your Secrets Store and your D1, under a master key that never leaves your Worker |
| Your files | Your R2 |
Every one of those is fetched live through a credential you minted, rendered, and discarded.
Stored
| Ours | Why |
|---|---|
| Our users and organizations | Somebody has to sign in |
| Our subscriptions | The dashboard bills for itself |
| Connection metadata | Project, environment, endpoint, and a reference to the credential |
| Alert settings | Each alert’s switch and threshold, the environments added to the schedule, the extra addresses you asked us to mail, and when each connection was last checked |
| What an alert already said | A SHA-256 of each occurrence’s key, kept 180 days, so the same failed renewal is not mailed twice. It names no purchase, person or amount |
| Privacy history, on Enterprise | Each change to the privacy inventory as a row in our D1, and each indexed inventory archived in our R2, sealed under a key of ours: field paths, class names and the build. Never a value. Kept for your retention window |
| Privacy settings | The check interval, the mail cadence, the environments watched, and the posture report’s switches |
| Saved reports, on Enterprise | Each report’s name, purpose, creator, the questions it names and its filter state, and a log of who ran it when, the newest 100 runs. Never an answer, and never a subject |
That is the list. The alert rows exist for alerts, and neither holds a row of yours. The report rows exist for saved reports: every run reads its answers live from your Worker, and they go with the screen or the file somebody exported. The privacy rows exist for privacy history, and are the one documented exception below.
The tempting exception
Caching your data for performance.
It is the obvious optimization, it would make every pane faster, and it is the thing that would quietly end the argument — because the moment a copy of your user table lives here, this has become the thing Pithy’s architecture exists to avoid.
So it is resisted. Where something genuinely must be held across a request, it is explicitly time-bounded and documented rather than left to accumulate.
What transits, and what happens to it
Panes are built to call directly from your browser to your Worker. A browser cannot hold a long-lived signing key, so the dashboard’s server mints a short-lived, single-scope, body-bound token and hands it to the browser, which calls your Worker itself. Response bodies on that path never touch our origin at all, and our private key never leaves our server. That path is waiting on the kit to answer a CORS preflight — the browser call path is the detail — and we have not built a proxy to stand in for it, because a proxy is the thing it exists to avoid.
The calls our own server makes are four, and none of them is a pane’s: the ping that proves a new connection answers, the key rotation that appends a successor, on Team the hourly reads alerts makes, and on Enterprise the scheduled reads of your privacy inventory. All four have to be ours, because only our server holds a private key and a schedule runs when no browser is open. Where a response does reach us, the commitment is a zero-logging policy on your data: response bodies fetched from your Worker are not logged, not cached and not persisted. An alert keeps the hash of what it already told you, and nothing from the response. The privacy series keeps the inventory itself, which names fields and never holds a value.
One kind of thing does come back to us: how an act ended. After a secret rotation, a data-subject export, an agent credential revoked when its member leaves, a repository’s tokens revoked, a leaderboard entry moderated, a game session closed or a result voided, a rating recalculated or reset, a player dropped from a matchmaking queue or an invite canceled, an act on one of your users or organizations, or a legal hold placed or released, your browser reports the outcome it already read, so the act lands in our audit trail. Each report is held to a strict shape: names, ids, closed states and the reason the operator typed. Never a secret’s value, a document’s contents, a score, a skill, or a person’s name or address, and nothing at all for any other kind of act.
Your Worker does not distinguish the two, deliberately: both are the same sixty-second, single-scope token, verified by the same code. There is no second token class to get wrong.
Why the architecture makes this cheap to keep
A promise you have to work to keep is a promise that eventually gets broken by a deadline.
This one is nearly free, because there is nowhere for your data to go. There is no Pithy-operated data plane — not for the dashboard, not for the kit, not at all. Storing your users would mean building something to store them in, which is a decision somebody would have to make on purpose rather than a corner somebody could cut in a hurry.
That is the difference between a policy and an architecture, and it is the reason this page can be a list rather than an assurance.
Verifying it yourself
Read the scopes you granted. pithy dashboard status --env prod prints them. The dashboard can do exactly what that row says and nothing else — your Worker enforces the row, not us.
Read your audit trail. Every control-plane call lands in it under its own actor kind, reads included — so what the dashboard did is answerable separately from what your users did. Reading your user table hands a management client every customer’s address; if only the writes were recorded, the trail would show one revoked session and say nothing about the customer list walked on the way there.
Revoke and watch it stop. pithy dashboard disconnect --env prod --yes --local deletes the row without telling us. Every pane fails immediately, because the authority was always your Worker.
The exceptions, stated plainly
Connection metadata is real data about you. Which projects you run, which environments they have, and what each Worker’s address is. That is not nothing, and it is not on the never-stored list.
On Enterprise, the privacy history is a map of your data model. Which of your fields hold which privacy class, build by build: field paths and class names, never a value, because your Worker never reports one. It is kept for the retention window your contract sets, thirteen months by default. When the plan lapses nothing more is read, the history stays read-only for thirty days, and then it is deleted. It is the one exception to the rule this page lists, and a second one would be a decision made on purpose, not a precedent this one sets.
Our own audit trail records what we did. Separate from yours, and about our own actions.
Billing details are held by our payment processor, not by us — the same arrangement any product using a merchant of record has.