The service
- There is no uptime promise. Free models change or disappear, and answers and speed can vary.
- Limits apply (see the docs). They can change without notice.
- The service can change or stop at any time during the beta.
Access and keys
- Access is by invitation. Each tester gets their own API key.
- Keep your key private. You are responsible for what is sent with it.
- Keys can be revoked at any time, for example after abuse or a leak.
What you agree to
- Do not use HybridAI for anything illegal, harmful or abusive, and do not try to get around the limits.
- Only send text you are allowed to share with a third-party AI provider. Do not send passwords, secrets, or medical or financial records.
- AI answers can be wrong. Check them before relying on them.
What is stored
- Waitlist: your email address, used only to contact you about access.
- API keys: a name, a short prefix and a one-way hash. The full key is shown once and not stored.
- Usage records: time, key name, route, model, status, time taken and token counts, kept for about 90 days.
- Rate limiting: IP addresses are used to count requests per minute and are kept for up to two days.
- Prompts are not stored by HybridAI. The one exception is saved answers: the reply to a plain prompt that has no private-looking data may be kept for up to 6 hours, under one-way fingerprints of the prompt, so an identical or very similarly worded prompt can reuse it. Send the header
x-hybrid-cache: offto avoid this. Prompts that look private, and requests that use your own keys, are never saved. - Model status counts: how many times each shared model worked or failed, never any prompt or reply.
- Voice tools: audio and text go only to Cloudflare's models and are not stored.
- Dev kit: the client files under /sdk are MIT-licensed. The gateway service and its code are not covered, and the names HybridAI and hybridai.cloud are reserved.
- Approval requests: the title and details you send (with private-looking values replaced first), the decision and the decider's note, kept for 14 days and then deleted.
Who can see your prompts
- Cloudflare runs HybridAI and its private edge models. The
private-edgeroute andedge-onlykeys keep a prompt on Cloudflare. - Other prompts go to Groq, and if that fails to Gemini (when set up), then OpenRouter and the free models behind it, then NVIDIA and Mistral (when set up). Gemini, NVIDIA and Mistral only receive prompts where nothing was redacted, because free plans at some providers may use prompts to improve their models; with your own key for them, your own account terms apply. Those providers handle your prompt under their own terms and may log it. Read their policies before sending anything you care about.
- In
redactmode, cloud providers receive placeholders instead of emails, phone-like numbers, card-like numbers, API keys and password-style values. Other personal details are not detected. - Detection is pattern matching and can miss sensitive text.
- When a listed free model is retired, HybridAI looks up current free models at Groq and OpenRouter and may use them. They are covered by the same provider terms as above, and they only receive what your key's privacy mode allows.
- Each response carries a signed receipt that lists every provider that received the request (see the docs).
Removal and contact
To be removed from the waitlist, or to have the usage records of your key deleted, write to hello@hybridai.cloud. We will act on it and confirm by email.