Privacy & data policy
Who operates this service
tiyuvta is operated by [LEGAL ENTITY — owner sign-off §12.8], founded and run by Avi Fenesh, Tel Aviv. Contact: [email protected].
What is logged
For every API request, tiyuvta logs request metadata for metering: token counts (prompt, completion, and cached tokens, itemized), timestamps, model id, request status, and the account that made the request.
On paid usage, prompt and completion bodies are not logged. There is no debugging exception: if a body was never written down, no investigation can read it back.
On free evaluation usage, prompts and completions are retained. What happens to them is stated under training, below.
Training
Paid usage is never used for training. Your prompts and completions are not used to train, fine-tune, or evaluate models — not by tiyuvta and not by anyone else — because they are not retained in the first place.
Free evaluation usage is training data. Prompts and completions sent on the free evaluation credit may be retained and used to train and evaluate tiyuvta's models. That is the price of the free tier, stated plainly. If you do not want your data trained on, use a paid key.
No third-party trackers
This website sets no third-party trackers, no advertising pixels, and no cross-site cookies. Any analytics, if added, will be privacy-respecting and self-hosted, and this policy will name them.
Data retention
Metering metadata (the fields listed above) is retained for 12 months for billing and abuse prevention, then deleted.
Subprocessors
GPU capacity may run on rented infrastructure (e.g. datacenter hosting providers). Those providers host the machines; they do not receive your request contents beyond what transits their network, and requests are TLS-encrypted in transit.
Your requests
To ask what is held about your account, or to request deletion, email [email protected]. For paid usage, request bodies are not stored, so deletion requests concern account and metering records. For free evaluation usage, you can request deletion of your retained prompts and completions — data already baked into a trained model cannot be extracted from it.
Changes
Changes to this policy are announced with dates. This page's history is public in the site repository — the policy itself has receipts.