Private by design
Local only uses no network or tokens and never switches silently to cloud. You decide when a request leaves the device.
Hybrid AI · private · transparent
SabettaLocal AI combines a model that lives on your iPhone with controlled access to selected online models. The device protects and trims context; Sabetta uses only the level of capability the task needs.
01 — The principle
Local is the starting point
Chat, voice, approved memory and encrypted documents can remain on the iPhone. In Smart savings or Maximum quality, the app selects only useful context locally, redacts recognisable sensitive data and sends the request to the Sabetta gateway. No OpenAI or Anthropic key is distributed to the phone.
Local only uses no network or tokens and never switches silently to cloud. You decide when a request leaves the device.
Preferences and lessons become active only after your approval and remain editable or removable.
Every online answer shows its route, model, tokens, cache and estimated usage. No hidden capability and no opaque consumption.
02 — How it works
One assistant, three levels of capability
Chat, voice, memory and documents work on the device. Choose it for privacy, offline continuity and no cloud usage.
The iPhone trims context and the Sabetta router selects the least expensive cloud model suitable for the request.
Context is still minimised, while Sabetta prioritises the most capable model for tasks that require it.
Approved memory, encrypted documents, voice, Siri, Shortcuts, Spotlight and the Action button make the assistant personal.
03 — Hybrid core
The right capability, not the most expensive by habit
The Sabetta router applies controllable rules to complexity, risk and recency. It can use an efficient cloud model or escalate to a more capable one. Sabetta does not resell access to a brand: it orchestrates different capabilities as one service.
04 — The pilot
Measure first. Promise later.
Version 0.9 is being distributed privately through TestFlight and the multi-provider gateway is live. The pilot will measure quality against direct cloud use, cost reduction, privacy, stability and willingness to pay. Plans and pricing will be based on real usage, not optimistic assumptions.
What we state precisely
SabettaLocal AI 0.9 is now in private alpha on TestFlight. We are preparing a small pilot to turn a working architecture into a simple, sustainable and transparent service.
Apply for the pilot