Start Here
How We Handle Your Data
Who can see your prompts, what we scrub from the traces we store, and how to get custom terms
camelStream is cheap because prompts and responses may be retained and used to train AI models, by us or by the providers we partner with, and may be used to select and measure ads and offers. Those arrangements are what keep the price flat. Here is who can see your prompts, how we scrub sensitive data, and what is never shared. The camelStream Terms and Privacy Policy are the binding versions.
Who can see your prompts
A prompt goes from your client to camelStream, then to the model provider serving your stream, and the response comes back the same way.
- The provider serving your request. A model has to read the prompt to answer it. The request reaches the provider without your name, email, camelStream API key, or IP address attached. Providers may retain prompts and use them for training under their own policies.
- Us. We store agent traces on our own infrastructure so we can debug the service, understand what agents actually do, and, potentially, train or evaluate models or select and measure ads and offers. Every trace we store runs through the privacy filter described below, and we separate it from your account before using it for anything beyond running the service.
- Partners in the arrangements that keep the price flat. Model-development partners may receive API content for training. Advertising partners may receive content separated from your identity, signals derived from it, or aggregate results. None of them receive your account details.
We never sell your data to data brokers.
What we scrub from the traces we store
Agent traces pick up things that should not sit in storage. An API key in tool output, an email address in a CSV the agent read, a phone number in a support ticket it summarized. Every agent trace we store runs through OpenAI's Privacy Filter, an open-source model built to detect and mask personally identifiable information in text.
We run the filter on our own hardware, so traces never go to a third party to be scrubbed. It reads the whole trace, including system prompts, tool calls, tool results, and model responses, and masks the following.
| What gets masked | For example |
|---|---|
| Names | A customer named in a support ticket |
| Email addresses | A personal inbox in a contact list the agent read |
| Phone numbers | Any format |
| Physical addresses | A street address in a form the agent filled |
| Account numbers | Bank, card, and customer account identifiers |
| Private URLs | Links that identify a person or a private system |
| Private dates | Birthdays and similar dates tied to a person |
| Secrets | API keys, tokens, passwords, and private keys |
The filter runs on the copy we store, not on the request itself. We do not scrub what you send on its way to the provider. The model has to read the original prompt to answer it, so a key you paste into a prompt, or one an agent reads off your machine and sends in tool output, reaches the provider like everything else. That is the same exposure as calling the provider yourself, minus your identity. No filter catches everything, which is why the Terms still ask you not to send data you are not allowed to share.
What is never shared
Your account details are separate from your API content. We never attach your name, email, billing information, or camelStream API key to a request, never share them with a provider, partner, or advertiser, and never use them for training or to select ads. Card payments are handled by Stripe, and camelStream API keys are stored hashed.
If you need custom terms
There is no opt-out on a standard subscription. Starting at 1,000 streams we can customize your terms so your data is never used for training or advertising. Your traffic goes only to providers that do not train on or retain data, and that can go into a contract if you want it in writing. Ask at camelai.com/stream/contact.
Last updated: September 8, 2026