Skip to content

Your client's data, and how to take it out ​

The question a developer's client asks them, not the other way round: if this goes away, what do I still have?

The short answer is that the two halves of a site are held differently on purpose, and both come out.

The site is a repository you already hold ​

Everything published lives in the site's own git repository, as files:

jtk/schema.jsonwhat the site is made of
jtk/pages/<path>.jsonevery page's published text
jtk/entries/<collection>/<slug>.jsoncollection entries — one document each
<name>.<locale>.json beside each of themthe words of every other language
jtk/shared.jsonthe text that appears on every page, the business's contacts among it
jtk/modules/<module>.jsona module's settings — the diary's hours, its rules
jtk/site.jsonthe site's own settings, the currency among them

An older site, on contract 2 until it is moved, has jtk/catalogue.json, jtk/content/ and jtk/bookings.json in their place.

That is not an export format. It is the published state — a build reads those files and nothing else, which is why a build is reproducible a year later from the same commit. "Give me this client as they were in March" is git checkout, and parting company is you keeping a repository that already has everything in it.

A site written from a chat, with no developer and no repository of its own (get_site → code: platform), is the business's code all the same: the panel downloads it as a zip at any time (The site → Build, We hold this site's code → Download as an archive (.zip); GET /v1/sites/{site_id}/code), pull_code brings it to a machine, and a GitHub repository of theirs takes it whole, with its history, from the same screen. Never tell a person the code cannot be taken out.

A collection also comes out as a spreadsheet a person can open, and goes back in from one — the panel:

GET  /v1/sites/{site_id}/collections/{name}/export
GET  /v1/sites/{site_id}/collections/{name}/sample                            the same columns, two example rows
POST /v1/sites/{site_id}/collections/{name}/import   {"csv": "…"}            the dry run
POST /v1/sites/{site_id}/collections/{name}/import   {"csv": "…", "plan": "…"}  the plan, written

The enquiries, bookings and orders come out whole ​

Everything a visitor sent or bought — the part that is not in git, and the part a business would actually miss:

GET /v1/sites/{site_id}/enquiries/export
GET /v1/sites/{site_id}/bookings/export
GET /v1/sites/{site_id}/orders-export

Three things are true of all three, and they are properties rather than promises:

  • Everything, not a page of it. The inbox stops at two hundred rows because a screen does; an export that stopped anywhere would be a backup with a silent hole in it.
  • No live subscription is required. Ever. Export behind a paywall is hostage-taking. When a site is frozen for non-payment, this is on the list of things the freeze does not touch.
  • A spreadsheet a person can open, and the same rows structured, in one answer. The owner of a nail salon reads it in Excel; a script reads the JSON.

They are in the panel as a button, on the enquiries, diary and orders screens. There is deliberately no MCP tool for any of them: an agent builds the machine and does not read what the machine collected, so taking the data out is something a person does, signed in as themselves.

The live orders, like the diary, come out only to whoever sees them in full: the business, or the studio while it holds the site as the business's proxy or through a window the owner has opened on Orders. A studio without one is refused the file, since it carries the buyers' names and addresses, and takes out only its own test orders.

The orders' VAT. When the shop charges VAT, the orders' file gains two columns after the tracking number: vat, the VAT of each order, and vat_rates, its rates — 8.1% 2.91; 2.6% 0.61 — as Stripe charged them. They are there only when an order in the file was taken with VAT on; a file of orders taken without it has no column of VAT. An order Stripe charged no VAT on says 0.00 and no rates; one whose rates could not be read has its VAT and no rates; one whose VAT could not be read has both cells empty. The amounts are in the site's currency; what a buyer paid in their own currency is on the order's page and not in the file.

As it happens, later. A business's own till or accounting tool will be able to hear each booking, payment and enquiry as it happens, as an event (Events). The events are written down today and delivered to nobody: webhooks, and the keys a business makes for its own systems, come later. When they do, an event carries ids and facts by default — a customer's name or contacts only where the business turns that on for that receiver, naming who receives it — and an event row on the platform holds no person and is kept 35 days.

The CSV files ​

Every spreadsheet the platform hands out — the enquiries, the bookings, the orders, a collection — is written the same way, and the collection's import reads it back:

  • A semicolon between cells. It is what Excel in Switzerland, Germany, Austria, France and Italy splits a file at when it is opened with a double-click; a comma file opens there as one column. An English Excel shows a semicolon file as one column — open it through Data → From Text/CSV; Numbers and Google Sheets find the semicolon themselves.
  • UTF-8 with a byte-order mark, so Excel reads ü, é and Cyrillic as themselves; CRLF between rows; cells quoted as RFC 4180 says, where they hold a semicolon, a quote or a line break.
  • A cell that would run as a formula — one beginning =, +, - or @ — starts with an apostrophe, so a name a visitor typed as =HYPERLINK(…) stays words. A phone number keeps its plus behind it; a plain negative number is left a number.
  • Amounts with a point and no thousands separator — 1250.00 — the currency in the column's name (price (CHF)). A Swiss or English Excel reads that as a number; a German or French one keeps it as text, which is still the amount.
  • Dates and times as ISO 8601 in UTC (2026-11-12T09:30:00Z); a switch as yes or no.

The import reads what a spreadsheet saves back: a semicolon or a comma (whichever the header line holds), with or without the mark, CRLF or LF; an amount with a decimal point or comma, a thousands apostrophe, space or point (1'250.50, 1 250,50, 1.250,50) and the Swiss 45.–; and in the panel a file in UTF-8 or in Excel's Windows code page. It takes the apostrophe off a cell that had one put on. A date where an amount should be — what a German Excel can make of 12.05 — is refused, not read as a number.

How long it is kept, and what is deleted ​

keptdeleted
an enquiry90 days from arrivalon its own then; the owner deletes one sooner from the inbox
a bookingthree years after its date — it is a business recordon its own then, paid or not: a payment is the business's own, in its own Stripe account, which keeps the record of it. There is no button to erase one sooner — ask us
page viewscounts by day, page and source; no cookie, no address, no person—
a picture or clip in the media libraryuntil the site is archived; one deleted leaves the library at once and the bucket at archival, older builds still naming itwhat it carries about where and how it was made — a phone's location, the camera, the time, an editor's notes (EXIF, XMP, IPTC) — as it is uploaded, its pixels untouched. A file uploaded before November 2026 kept what it came with; uploaded again, the copy kept is clean
the published sitein the site's repository, which is yoursnever by us

Nothing is deleted because a subscription lapsed. A frozen site stops accepting new enquiries, bookings and orders — that is what the freeze is for — and everything already collected stays readable and exportable. Deleting the data of a site that has been lapsed for a year is decided and not built; until it is, lapsing deletes nothing.

Where it is held, and what it passes through ​

the platform's database — drafts, enquiries, bookings, accountsa managed PostgreSQL in an EU region, with point-in-time recovery
its off-site backupsBackblaze B2, EU Central (Amsterdam)
pictures and built sitesCloudflare R2, served by Cloudflare's network wherever the visitor is
the APIa server in the EU (DigitalOcean)

The services a visitor's data passes through on its way, each only when the site uses it: Cloudflare (serving the site, the bot challenge in front of a form, page counts), Resend (the letters to the visitor and the owner), Telegram (when the owner or the visitor connects it), Stripe (when a booking or an order is paid online). With VAT on, the name and address a parcel goes to are also kept as a Customer on the business's own Stripe account, which Stripe Tax reads; Stripe's dashboard has them, and they are the business's there. The people who sign in to the panel do it with Google or an emailed link.

Who is responsible for what ​

The business is the controller of its visitors' data: it decides to collect an enquiry or a booking and what to do with it. The platform keeps and delivers that data on the business's behalf — a processor — and the studio stands between them as the one who set the site up.

What this page does not tell you yet ​

Things a client's lawyer asks that are not answered here, and saying so is better than implying an answer:

  • A data-processing agreement you can hand on. There is none yet.
  • The availability commitment.
  • An accessibility statement for the sites the scaffold builds. They have not been audited against a standard.

These are commitments rather than descriptions, and this documentation will not invent them.