Translate anything. Keep nothing.

Wiring translation into a product usually means handing your users' text to an outside vendor through a third-party API. Teddy is the privacy-minded alternative: the same one-endpoint integration, but the text goes to a privately-run model instead of a commercial translation provider. It's still a hosted service a client connects to, the same as any other API. The difference is who's on the other end, and what happens to the text once it arrives.

The call itself is deliberately plain: give it text and a target language, source is optional and gets auto-detected if left out, and the translation comes back synchronously. No login flow, no session, just an API key on every request. The in-browser translator anyone can click through is a thin convenience layer on the same endpoint every other integration calls.

Confidentiality here isn't a policy promise, it's structural. Translation text, in and out, is never stored. What gets recorded is usage only: request counts, word and token counts, language pairs, timestamps, enough to meter and audit without retaining a word of anyone's actual content. Requests are served by a privately-run language model rather than routed out to a commercial translation vendor, so client text never leaves for an outside AI provider in the first place.

Rolling this out across more than one integration is where the key model matters. A client gets one master key, and mints scoped sub-keys underneath it per app or per user, each seeing only its own usage and each independently revocable. One compromised or retired integration means rotating one key, not resetting everyone else's credentials too.