Local-first protected storage
The private-beta architecture keeps the personal vault and encrypted documents on the user’s Mac. Protected data is encrypted with per-user keys, with the local wrapping key stored in macOS Keychain where supported.
PRIVACY & SECURITY
An assistant this useful has to earn trust.
HeyJim handles context that ordinary chatbots often never see. The product is therefore being built around minimisation, explicit permissions, protected storage and action confirmation.
Least-context AI
Ordinary questions should not send an entire personal vault to an AI provider. HeyJim is being built to retrieve a small relevant set of facts and keep secret values out of routine prompts.
Real-target confirmations
For browser and desktop actions, the safety decision should be based on the actual resolved control, form or URL rather than whatever label the model guessed. Higher-impact actions pause for confirmation.
Protected values are different
Passwords, signatures, passport and banking details, NI numbers and other high-sensitivity information should not behave like ordinary memory. Their use is restricted by task and destination.
Connected services stay visible
Email and other integrations are explicit connections. Users should be able to see them, disconnect them and remove stored OAuth credentials.
History without leaking secrets
Action history can record what happened while scrubbing protected values instead of turning the audit log into another sensitive-data store.
What “local” does and does not mean.
The vault can be stored locally, but HeyJim is not completely offline. When you use AI-powered features, selected task context may be sent to the AI provider configured for the product. Connected email features communicate with the relevant provider. The privacy notice should always describe those flows clearly.
Read the privacy notice →