Claude in Canada: Where Does Your Data Go?
Use Claude through a Canadian service. Know what still reaches Anthropic.
Updated
You can use Claude through Rideau AI, a Canadian-owned service. Your conversations and files are stored on hardware we own in Ontario. When you choose Claude, Anthropic processes the request. Turn on Redact to replace identifying details before sending.
Paying a Canadian company, storing your history in Canada and running the model in Canada are three different things. Here is what each means when you use Rideau AI.
What stays in Canada when I use Claude?
Your conversations and files are stored here, and our safety scanners, redaction and PDF processing run here too. Claude is an outside model: selecting it involves Anthropic, even though your conversation history is still stored with us.
The model picker shows who runs the model you select, where it’s processed and what its provider keeps. A Canadian subscription doesn’t mean the model runs in Canada. See how we label every model.
What the label says today
The model picker labels Anthropic’s Claude models this way:
| Operator | Hardware | Processed in | Provider keeps |
|---|---|---|---|
| Anthropic (US) | Multiple vendors (US) | Global | Up to 30 days |
“Multiple vendors” means Anthropic names several US cloud companies that run its models and doesn’t say which one handles a given request. “Global” means the request can be processed in any region Anthropic uses. Some models carry their own exceptions, such as a longer retention period for automated safety review. Each model’s label shows its own terms, and the picker always shows the current one.
What does Claude receive?
| Your choice | Who answers? | What the outside model receives |
|---|---|---|
| Claude with Direct | Claude, through Anthropic | Your request and the context sent with it, without redaction. |
| Claude with Redact | Claude, through Anthropic | A version with identifying details replaced by placeholders, together with the rest of the request. |
| Private | Our AI in Ontario | No outside AI provider receives your message or documents. This uses our model, rather than Claude. |
In Redact, we restore the original details in the answer. Redaction isn’t perfect, but you can always preview before sending. When the whole idea is sensitive, keeping the AI processing on our hardware may matter more than the model you would otherwise choose.
Can Claude help with documents without reading the originals?
Consider a services agreement and an NDA. You want to compare their confidentiality terms and follow citations to the relevant pages. You may not need Claude to know the parties’ real names or addresses to do that work.
We read the PDFs on our hardware. In Redact, the same identifying detail keeps the same placeholder across documents. Claude can follow those connections. If it needs context hidden by a replacement, it can ask our AI to read the originals here; we redact that answer before returning it.
You still need to check the source documents and the answer. The document examples show how that looks in the app, and Can I Use AI on Confidential PDFs? covers each mode.
Are training and retention the same thing?
No. A provider can avoid using requests to train its models and still retain them for a period. Anthropic’s commercial policy says it does not use API inputs and outputs for model training by default. Its retention policy describes storage periods and exceptions. These are separate questions. Read its commercial training policy (opens in a new tab) and retention policy (opens in a new tab).
We use the commercial API arrangement. No model we offer trains on your conversations. The model’s label tells you what the provider keeps; this guide does not promise zero retention. Redact limits the identifying details you send, while the rest of the request remains subject to the provider’s terms.
What if no outside AI provider can receive the work?
Choose Private. Our AI answers on our hardware in Ontario. It’s a smaller model, so match it to the job. External tools have separate boundaries: Private disables web search, but a link you ask the app to fetch can still go to an outside retrieval service. See Where your data goes.
If you are still choosing a service, start with our Canadian ChatGPT alternative guide. If you are evaluating customer information for business use, the PIPEDA guide explains the questions to bring to that review, whichever model you prefer.