Skip to content

Trusting the answers

Where does my business data go when I connect my AI?

A simple, honest guide for owners who want to know where their data actually is after they connect an AI, not a reassurance that avoids the question.

Concept guideApplies to The Edge or your cloud tool · Claude

Short answer

Two things are true, and the honest answer names both. Your system of record, The Edge or your own cloud tool, stays where it runs. We do not move it or replace it. When the connection uses a scheduled export, a one-way export of the data lands in a copy we host, an analytics database we run, and every question the AI answers reads that copy, never your live system. When the connection is a live API instead, there is no separate copy, and the AI reads the vendor’s cloud directly. Either way, this is a read path, and any write-back is a separate, deliberate setup, off by default.

An owner deciding whether to connect an AI, who wants a straight account of where the data lives afterward, not a reassurance that avoids the question.

  • To know, per system, which connection type it uses: a scheduled export or a live API. The answer is different for each.
  • A short list of which systems you are connecting.
  • The privacy questions that matter to you, so we can answer the specific ones before anything goes live.
  1. Your source system stays put. The POS, the books, the cloud tool: the system of record keeps running where it runs. Connecting an AI does not move it or swap it out.
  2. An export makes a copy, and we name where it lives. If the connection is a scheduled export, a one-way export of the data leaves the source and lands in a copy we host, an analytics database we run. That copy, not your source system, is what every question reads. What it contains is what the export contains, and no more.
  3. A live API makes no copy. If the connection is a live API, there is no hosted copy. The AI reads the vendor’s cloud directly, through the gateway, as of the moment it asks.
  4. The AI reaches the copy or the API, never the source machine to change it. In both cases the AI talks to the gateway, and the gateway talks to the copy or the API. Any write-back is a separate, deliberate setup, off by default.

It can

  • Keep your system of record in place, running where it always ran.
  • Tell you, per system, exactly where the data the AI reads lives: a copy we host, or the vendor's cloud.
  • Limit an export's copy to what the export contains, nothing wider.

It cannot (honestly)

  • Honestly claim nothing moves. With an export, a copy does move, and we say so plainly.
  • Answer the specific privacy questions in the abstract. Who operates the copy, how long it is kept, and who else touches it are real answers we give per deployment.
  • Read anything the connection was not set up to reach.

Your system of record does not move; it keeps running where it runs. What moves depends on the connection. A scheduled export sends a one-way copy of the data into a copy we host, which the AI reads. A live API sends no copy; the AI reads the vendor’s cloud in the moment. Anyone who tells you “nothing moves” is skipping the part that matters, which is exactly the part this guide names.

Who runs the hosted copy, and how long is it kept?

Section titled “Who runs the hosted copy, and how long is it kept?”

These are real questions with real answers, and we give them per deployment rather than in the abstract. Who operates the hosted copy, how long data is retained, and which subprocessors are involved are confirmed for your setup and settled before anything goes to production. We would rather answer them precisely than wave at them.

EvidenceSupported capability under gate

The connection in production on The Edge uses a one-way export into a copy we host, which every question reads instead of the live system. We last verified it running on August 27, 2026. That is the model. The specific facts a careful owner asks next, who operates the hosted copy, how long data is retained, and which subprocessors are involved, are confirmed for your deployment and settled before anything goes to production.