Take everything with you

Nothing you build here is locked in. If you leave, or just want a copy, Take everything on Settings, Company packs the whole account as you read it into one zip. Your data stays where it is; this is a copy, as of the day you take it, and it never changes anything. To have a copy kept for you every night instead, in a bucket of yours, see Your data, in your own bucket: a zip by hand when you want one, or the whole warehouse in your bucket each night.

What is in it: every table you may read as a Parquet file (and as CSV too when it has 100,000 rows or fewer), with a manifest listing each table's rows and column types; your models as the same dbt project Download as a dbt project gives (and each Python model's code, the saved queries and the metrics as files); every kept file, every version of it, under files/<name>/<version>/, and the original behind each table loaded from a file; and each dashboard as a folder: its page code as it is stored, a snapshot of every number and saved query the page reads, and an index.html that opens the dashboard from the folder in a browser, read-only, drawn from the snapshot with a copy of the runtime that is in the zip. A README at the root and in each dashboard's folder says how the pieces fit.

How it is made

Press Take everything. It is built in the background, however big the account, so nothing times out: the card under the row says how far it has got, you can leave the page, and your phone (if notifications are on) and your email hear when it is ready. Back on the card, Download saves the zip; the link it uses works for ten minutes and the card gives a new one each click. The file is kept for a day, then deleted; take it again after that. One at a time: pressing the button while one is being built shows that one.

You get what you can read, and only that. A dataset you see masked comes out masked; a file you may not open is listed in the manifest as closed, with why, and left out. Someone in a group other than Everything cannot take it: their data is what is shared with them. The zip's bytes count toward the day's data served, like any download.

From a terminal or an agent

ovs take-everything --out ~/acme-everything       # unpacked into the folder; start with its README.md
ovs take-everything --out acme.zip               # the zip, whole

The command waits while the warehouse builds it, then saves it. Over MCP the tool is take_everything: with no id it starts one and answers its id; with the id, where it stands and, once ready, the link. The door behind both is POST /everything, followed with GET /everything/<id>.

Opening a dashboard from the folder

Open dashboards/<name>/index.html in a browser. The page draws itself exactly as it was built, from the snapshot in its data/ folder, with a bar across the top saying it is a read-only copy and the day it is from. What needs the app is refused on the page rather than failing quietly: marking a row, filling in a table, downloading a table from the page, and where a number comes from. A dashboard that works a question out as it runs (a period from its own date picker) can ask for something that was not on its list of numbers when the copy was made; the page says so where that number would be.