Guides
Data and privacy
What we record about each request, what we don’t, how to ask for zero data retention, and what you need to tell your own users.
On this page
What we record
For each request, we keep one row in your usage history, so you can see what it cost. It holds:
- the request ID, when it happened and how long it took;
- the model, and which API key sent it, with the key’s name at the time;
- whether it succeeded, failed, was stopped or wasn’t run, with an error code if it didn’t finish;
- its input, output, cache and reasoning token counts;
- what it cost, and the prices used to work that out.
The usage record never contains the words of your prompts or the replies. Our request logs don’t either: they hold counts and IDs.
You choose how long the history is kept, from 7 days to for good, and can delete it at any time. See How long history is kept.
Your prompts and replies
We don’t store the content of API requests or replies. To produce a reply, your request is handled in memory by Kalsenti, checked by our safety filter, and sent to the host that runs the model you chose. The reply comes back the same way.
- The model’s host processes your request to answer it, and may keep it for a time. The labels on the Models page say where each model is hosted, how long its host may keep data, and whether the host may train on it. See Hosting and data labels.
- Kalsenti doesn’t use your API requests to train models.
- Your requests are readable while they’re being answered, by us and by the model’s host. Bear that in mind when deciding what to send.
Safety filtering
Every request, on every model, is checked automatically for a small number of serious categories: the messages you send, including earlier turns and any document text in them, and your tool definitions. A blocked request isn’t sent to a model. Replies are only checked in private chat on the Kalsenti site, not on the API. We record the category, the rule that matched, the language and the time against your account for 30 days, never what was written. Several blocks in a short time pause chat and the API for your account for an hour. See Blocked messages for the error codes.
Telling your users about AI
If you build Kalsenti into your own product, our Terms of service and Acceptable use policy ask you to:
- tell your users when they’re talking to an AI, or seeing content generated by AI, where the law requires it;
- make sure your users follow our Acceptable use policy, with terms of your own for them that are at least as strict;
- give your users a way to report abuse, act on what they report, and tell us promptly about serious misuse;
- help us work out which of your users sent a request, if we ask;
- give your users your own privacy notice, covering the data you send us.
The terms are what apply, so read the whole section before you launch.
When the requests you send hold personal data about your users or customers, our Data processing agreement covers how we process it for you, as part of the Terms: you decide what is sent, and we act on your instructions. It applies without a signature. For a signed copy, or the list of subprocessors, email privacy@kalsenti.ai.
Policies
| Policy | Covers |
|---|---|
| Terms of service | The agreement for using Kalsenti, including the API. |
| Privacy policy | What we collect, why, who we share it with and how long we keep it. |
| Data processing agreement | How we process personal data on a business customer’s behalf. |
| Acceptable use policy | What you and your users can’t use Kalsenti for. |
| Refund policy | When a purchase can be refunded, and how to ask. |
Questions about any of them, or about your data, go to support@kalsenti.ai.