Why local-first matters for AI assistants

Separate local storage from external requests

"Local-first" can describe several architectures, so the useful question is which operations and data stay on the device. Noet stores indexed data in a local SQLite database on the Mac. Retrieval starts locally, while selected context may be sent to configured AI providers for synthesis.

That distinction matters. Local storage can reduce the amount of indexed work data held in a separate cloud search database, but it does not mean every feature is offline. OAuth providers, optional sync and AI requests still have their own data paths. The Security and Privacy pages describe the current boundaries and subprocessors.

Review an account before connecting it

Before connecting a work account, review the requested provider scopes and your organisation's policy. Check which feature initiates an external request, which data is selected and whether the action requires confirmation. Local storage is one control, not a substitute for access control, encryption or clear retention rules.

Try a question using information you can safely verify, then inspect the sources. Adding more accounts increases the context available to the assistant, but does not make every account necessary for every task. Keep services disconnected when their data is not needed for your workflow.

Treat screen capture as a separate choice

Moments is optional screen recall. Its pause, exclusion and retention controls let you choose which activity becomes searchable. Exclude sensitive applications and window titles before enabling capture, and revisit those choices when your work changes. A local index still needs sensible device access and retention controls.

Verify the current data boundary

Noet's public security documentation is the source of truth for the product's current design. This article does not claim that local-first architecture alone satisfies a regulated organisation's legal or compliance obligations.