Contents
Most AI providers promise not to train on your data. That’s one question out of seven. A prompt can still be logged, kept for 30 days, read by a reviewer, passed to another company, or stored as chat history on a server. Here’s what to ask, and what each answer looks like at SolCard Inference.
1. Are prompts and answers logged?
Logs are where data leaks by accident. A provider can promise no storage and still write request bodies to a debug log, an error tracker or a monitoring tool.
Ask what a log line holds. The right answer is a list of fields, with the text of your message not on it.
SolCard Inference: our logger writes from a short list (which account, which model, how many tokens, how long it took) and drops everything else, so a message can’t reach a log even by mistake. Automated checks fail the build if code tries to log one.
2. How long is anything kept?
“Deleted after 30 days” is a retention policy, not zero retention. For those 30 days, your prompts sit in a database that can be breached, subpoenaed or read.
Ask for the number, per kind of data.
SolCard Inference: messages and answers: never kept. The billing line per request (time, key, model, tokens, cost) is kept. So is your account’s email or wallet address.
3. Does the model host keep it?
Many APIs are a front end on someone else’s models. The front end’s policy doesn’t bind the company running the model.
Ask who runs the model, and what that company keeps.
SolCard Inference: our model host runs open models on its own hardware, and our account there is zero data retention: your message is processed in memory and not stored. The host isn’t told who you are, either: it never gets your name, email or IP address. If it’s ever unreachable, the backup route is set to zero retention only, and refuses any model that would keep your message.
4. Can a person read it?
Some providers keep flagged conversations for human review. It’s worth knowing whether your request could be one of them.
SolCard Inference: every message passes an automated content check that records its decision, not your text. No person reads your message.
5. Where does chat history live?
An API call can be zero-retention while the same company’s chat app stores every conversation on its servers.
SolCard Inference: your chats are stored in your browser, on your device. Turn on Sync and they’re end-to-end encrypted before they leave it: our servers keep a copy they can’t open, and only your devices can.
6. What leaves the zero-retention path?
Web search, file uploads and plugins often go to other companies with their own rules. A good provider tells you when that happens.
SolCard Inference: web search is off until you turn it on, and the app tells you the moment you do: the query goes to a search engine, which keeps it by its own rules. Files you attach are read on your device and sent as text with your message; no upload service holds them.
7. Can you check any of it?
Promises are cheap. Look for something you can verify yourself.
SolCard Inference: open the chat app with your browser’s developer tools on the Network tab. Every request goes to app.solcardinference.ai or api.solcardinference.ai, and the page’s content-security-policy header keeps it that way. No analytics, no trackers.
The checklist
| Question | What you want to hear |
|---|---|
| Logged? | Metadata only, never the text |
| Kept how long? | Messages: never |
| Model host keeps it? | No, in writing, for the provider’s account |
| A person reads it? | No |
| Chat history? | On your device, or encrypted so the server can’t open it |
| What leaves? | Named, off by default, flagged when you turn it on |
| Can you check? | Yes, in your own browser |
Keeps nothing follows one message through SolCard Inference, start to finish, if you want the whole trip.