Your data, plainly.
We cache verified answers, not your prompts. Most security pages are written to make you stop asking. This one answers the question, then tells you how to check.
What we never do
- We never keep your prompts or responses in your usage records. Your history holds token counts, timings, and costs. Not text. You can ask us to prove it.
- We never share anything "similar." Other caches reuse an answer because your question looked ninety-five percent like someone else's. Similar is not correct. A Tīrtha cache hit is exact and verified, or it is not a hit.
- We never confirm your resources exist to anyone else. If another account asks for your ticket, your thread, or your key, they do not get "access denied." They get "not found." We do not even reveal there is something to deny.
- We never train on your data, and no provider trains on it silently through us. Our primary serving provider's no-training terms are verified in writing. Any provider that cannot meet that bar does not serve your traffic.
What is standing guard right now
- A web application firewall in front of everything.
- A spending cap on every key, on by default, so a leaked key cannot quietly become a huge bill.
- Cache entries addressed by a content hash computed on our servers. A crafted URL cannot trick the cache into serving someone else's data. That is the exact bug class behind the famous cache leaks you have read about, and it cannot happen here by construction.
- Per-request token metering you can see, so the bill is never a story. It is arithmetic.
What you control
Today
Your keys, revoked any time. Your spend, capped and enforced on our servers. Your participation: the beta is invite-only and opt-in from the very first touch.
Coming, in this order
A per-key switch so your answers never enter the shared cache. Spend caps you set yourself. A view of exactly what we hold about your account. One-click export and deletion.
Every control we add follows one rule: it can only reduce what we hold, or show you what we already do. We do not add levers that reach into your data.
Do not trust us. Check us.
Every claim on this page is written to be checkable, and we would rather show you than promise you. Email [email protected] and ask how any of this works. A real person answers, and if we get something wrong we will say so in public. Last week we caught our own best benchmark result being false and published the finding the same day. That is the standard we hold ourselves to, so you can imagine how we treat your data.