AI & Data Use Notice
Last updated August 31, 2026 · version 2026-08-31
Galaxy Copilot, receipt scanning and voice use artificial intelligence. This notice says what that means for your data, what the AI cannot promise, and what stays your decision.
What the AI features do
Copilot answers questions about your business by reading the records already in your workspace, drafts messages and documents, and proposes actions. The receipt scanner reads a photo and extracts the vendor, date, tax and total. Voice turns what you say into text and speaks the answer back. Under the hood these use large language models, computer vision, optical character recognition and speech recognition from third-party providers, and we may change the model or the provider without notice.
Where your data goes
When you use Copilot, your question, the conversation, the workspace information the assistant looks up and any photo you attach are sent to an AI provider outside Galaxy Unit so the answer can be produced. Voice audio goes from your browser directly to the provider and never passes through our servers.
Our providers today are Anthropic (text and photos) and OpenAI (voice and speech), both in the United States. If you store your own provider key in settings, your requests go to your own account with that provider instead. The full list is on the Subprocessors page.
If none of this is acceptable for your business, do not use the AI features. The rest of the product works without them.
Your data trains nothing
We do not train any model on your data, and no such capability exists in the product. Your records are used to answer your questions in your workspace — that is the service you switched on — and never to answer another business’s. If we ever want to use identifiable data to improve a shared model, that will be a separate, clearly worded, opt-in choice, never a sentence buried here; and we will not pretend a business can consent on behalf of its own customers when it cannot.
We may use figures that identify no one — how often a feature is used, how often an extraction is corrected, how long an answer takes — to run and improve the product. Removing a name is not what makes data anonymous; we hold ourselves to a higher bar than that.
It can be wrong
AI output can be inaccurate, incomplete, out of date, invented, biased, or simply based on a wrong record in your workspace. A confident, polished answer does not mean a human checked it — no one at Galaxy Unit reviews your answers. Scanned receipts can misread a digit or put a value in the wrong field. Suggested prices, rankings and follow-ups are estimates, not facts. The same question can produce a different answer twice.
You review before you rely
Before an AI output reaches a customer, a tax authority or a bank, check it: amounts, taxes, names, dates, recipients, payment details. Where Copilot proposes a change to your data, it asks you to confirm first; the confirmation is yours, and so is the result. Where you knowingly switch on an automation that acts without asking each time, you are choosing that, and the rules for automations in the Terms apply.
Not professional advice
Galaxy Unit is not your lawyer, accountant, tax adviser, bookkeeper of record, financial adviser, insurer or engineer. Tax estimates, expense categories, bookkeeping views and drafted contract language are management tools and drafts. A detailed answer does not create a professional relationship. Where your situation warrants professional advice, get it.
The decisions stay yours
You remain responsible for the prices you quote, the invoices you send, the taxes you register for and remit, the messages you send your customers, and every business decision — whether or not an AI suggested it. Relying on an output does not move that responsibility to us. Nothing here removes a responsibility the law does not let us exclude for our own conduct.
Your controls
You can stop using Copilot at any time, clear its memory of your preferences in settings, use your own provider key, and delete your account — which erases the workspace the AI read from, including derived indexes and memories, on the schedule in the Privacy Policy.
