Journal · 4 November 2026
What local-first AI means for Sabine
Local-first is a design choice: keep important functions close to the user, with clear boundaries for anything that leaves the device.
More control over data
Sabine is being designed as a locally deployable system for research, memory and task execution. The goal is to let you decide which data is available, which tools may be used and when human approval is required. This is a development goal, not a claim that every feature is finished.
Local does not always mean offline
A local model can process text on the device itself. Current web research needs an internet connection. External model services may also receive information if you deliberately enable them. The boundary between local processing and external services should therefore remain visible and understandable.
The model router as a boundary
The Neural Core design separates model choice from the rest of the system. Existing open-weight models provide a temporary foundation during development. Our long-term goal is our own language model once resources, capacity and measurements support it.
Privacy also needs operational care
Running locally does not automatically guarantee privacy or security. Permissions, logs, storage, deletion and external connections also need attention. Sabine aims to make those choices verifiable rather than hide them behind a broad promise.