Exploration
Measuring offline translation
We’re exploring offline translation and captions on top of streaming speech representation—this is a measurement spike, not a product release.
We’re exploring whether offline, local translation can sit cleanly on Tojito’s streaming speech representation—the same reusable layer that already feeds profanity filtering.
The current work evaluates captions-oriented translation as another consumer of representation: understand speech locally, then project something useful for a listener, without turning the mute path into a translation engine. That architectural fit matters as much as any single model choice.
This is a measurement spike, not a product release.
We are not announcing translation availability, TTS, dubbing, packaging timelines, or quality targets. Offline translation and live captions remain on the roadmap as direction. Until a capability is ready to ship, Updates will describe exploration in those terms—and Changelog will record what actually lands in a release.
Local-first still applies: the interesting question is whether meaningful translation work can stay on-device as the platform grows. Measurement answers readiness questions; it does not invent a ship date.