Anthropic
United States (San Francisco)
- What it does
- Claude writes the replies in chat, texts, WhatsApp and email and reads the business’s website for businesses on US models. On the other settings it is the fallback when the DeepSeek route fails or is slow to start, and it answers any message that includes a PDF. Where Amfion configures it for an account, it can also handle the lookups behind phone calls. For a law firm, whatever its setting, it also checks the assistant’s replies for anything that reads like legal advice after they are sent, including the assistant’s side of a phone call once the call has ended.
- Data it receives
- Each message and the conversation so far; the business’s instructions and published facts; the results of lookups and actions (availability, booking details, and the names and contact details a person gave); photos and PDFs sent in a conversation; the text of the business’s website when it is read. For a law firm’s reply check: each assistant reply, or each of the assistant’s turns of a call transcript, with the message it answered (a short reply with nothing legal in it is cleared without being sent).
- Where it processes
- Anthropic stores data in the United States. Its default routing may process requests in the United States, Europe, Asia or Australia, so US processing is not guaranteed.
- How long it keeps data
- Up to 30 days, for example to monitor for abuse; content Anthropic flags under its usage policy can be kept longer.
- AI model setting
- Recommended · US-only AI providers · US-hosted open models