ARCY AI
SettingsInstall

Bring Your Own Data

Bring the people, events and text your tools already hold, map their columns once, keep every other column, and get a Report about the rows that landed.

Settings > Install > Bring Your Own Data takes what your other tools already hold. Ten Datasets in three groups: People write attributes, Events become past events on their timeline, and Records are text ARCY reads as themes. Every import also gets its own Report (Reports).

Before you start

  • Only an admin imports or deletes an import; a member reads.
  • You can import before your ARCY call: one import and one Report until then.
  • Each import names one environment, Production by default.
  • One file of up to 100 MB with a header row: CSV, or XLSX read from its first sheet.

Setting it up

Pick a Dataset card; its dialog lists the columns it needs, with Download an example. A People or Events row names its person in user_id, email or anonymous_id, the user id winning; a Record needs none and keeps the person when it has one. A Support tickets file also reads its status, priority, channel, first response and resolution times, satisfaction score and reopen count; Reviews and Survey responses read the rating.

Upload and read reads the headers. Map the columns shows each header beside the field it goes to, matched by name or a common alias. A header no field claims opens on Keep with the row, or Keep as an attribute on a People Dataset. Fix a wrong one, drop one with Set aside, then Read the file; a required field with no column holds the button. The mapping is remembered per Product and Dataset and offered first next time.

Every column you keep lands with its row and is counted in the Report, so rename nothing before uploading. ARCY keeps up to 20 a row, 200 characters each.

The Import summary counts the rows, the people to create and match, what was skipped and why, and the columns ARCY kept. New columns appear under New attributes with a guessed type to tick, and new event names under New events; an Orders file proposes order_placed with each row's amount. Import declares what you ticked, then imports.

Your coding agent sends the same rows (ARCY MCP); its import waits in Past imports.

What it changes

  • An imported person appears on Users and Organizations. A person imported by email alone shows No user id yet until a verified identify() sends that email with a user id: they become that user, once. An anonymous_id joins a live visitor only through identifyAnonymous().
  • A blank cell never clears a value; the latest write wins. A repeated event, or a Record with the same id, is skipped: the same export twice changes nothing.
  • Imported events, Records and kept columns are not tracked events: the Report and the timeline read them, and Lift, Traffic, Conversations and the chat assistant never do (Data retention).
  • A card number, IBAN or API key is refused in a cell and masked inside a Record's text, and a column named like a password, a card security code or an IBAN is refused whether or not a field claims it (What we never do).

Delete this import removes its events and Records, and the people it created who never visited and were never linked to a user id. Attribute values it wrote stay. It cannot be undone.

When it goes wrong

  • Every row skipped as No Match key: no column maps to user_id, email or anonymous_id. Map one and read again.
  • Rows skipped as Email matches more than one person: two people carry that email. Give those rows a user_id.
  • Rows skipped as No text or Amount could not be read: that column is empty, or not a number.
  • The file will not open: save it as .csv or .xlsx from Excel or Google Sheets and upload again.
  • A mapping nobody confirms and an Import summary nobody imports expire after 30 days.

See also

On this page