Per-principal paths
Application operations are scoped to the current principal, with row-level security reinforcing those data boundaries in the database.
How Habit protects your data and your decisions.
For an AI assistant, privacy is a set of data paths, permission boundaries and action controls—not a promise that information never moves. Here is what Habit can responsibly say about those boundaries today.
Data boundary
Private starts with knowing where data goes.
When you connect a provider, relevant data can move between that provider, Habit’s application services, storage and selected AI processors.
Habit requests provider-authorised permissions for supported sources. The application selects context needed for a brief, response or proposal; privacy should be evaluated across that whole path, not inferred from a label on the interface.
Primary application data is stored in Supabase’s Paris region. A Paris region does not mean every processor or every processing operation stays in Europe. Cross-border processing and data-residency obligations remain questions to assess against the current provider path and legal requirements.
Stored data
Isolation and encryption have specific meanings.
Habit combines application-level scoping, database controls and encryption for sensitive stored fields.
Protected stored fields
Sensitive stored fields use AES-256-GCM encryption with a separate encryption key for each principal.
Separated credentials
Connected-provider credentials are addressed separately from ordinary application records so access paths remain explicit.
Encryption at rest is not the same as a system where only the user holds a decryption key. Habit’s current architecture is not described as end-to-end encrypted, because authorised application services need to process selected context to provide the service.
Operations
Use less content in operational logs.
Operational visibility should help diagnose failures without copying private messages into every diagnostic trail.
Habit’s logging approach favours identifiers, statuses and event types rather than the content payloads being processed. This reduces unnecessary exposure while retaining a path to investigate a failed job or an unexpected state.
Selected context may still be sent to configured AI providers when the product needs a model to prepare output. Claims about provider retention or model training must be verified against the applicable provider terms, account settings and contract. This page does not replace that evidence with a universal “never” claim.
Action safety
Protect the decision as well as the data.
An assistant can create harm without leaking a record. That is why action authority has its own boundary.
Review by default
Consequential actions are presented for approval so you can confirm, modify or decline the intended change.
Category-specific opt-in
Eligible automation requires a deliberate standing authorisation for that bounded action category; it is off by default.
Destructive changes stay gated
Calendar updates and deletes remain review-gated rather than inheriting authority from an opt-in to calendar additions.
Your controls
Permissions should be reversible.
Pausing assistance and disconnecting a provider are distinct from memory management, retention and deletion. Each control needs a defined, verified effect.
Disconnecting a provider stops future authorised access through that connection. It does not by itself establish what retained information is deleted. Retention and deletion behaviour require a verified process and accurate legal disclosure; no immediate or universal deletion claim is made here.
Read the published Privacy PolicyQuestions
Security questions worth asking directly.
- Is Habit end-to-end encrypted?
- No such claim is made. Habit encrypts transport connections and sensitive stored fields, while authorised services still need to process selected context.
- Where is data processed?
- Primary application data is stored in Supabase’s Paris region. Connected providers and selected processors may operate in other regions, so this is not an all-EU processing claim.
- Is connected data used for model training?
- That answer depends on the applicable AI provider, contract and account settings. No universal training claim is made on this page; verify the configured provider path and terms.
- Can Habit act without confirmation?
- Review is the default. Only eligible, bounded actions can run under an explicit standing authorisation; destructive calendar changes remain review-gated.
- What happens when I disconnect a source?
- Disconnecting stops future access through that authorisation. Retained data and deletion are separate matters whose exact product and legal handling must be verified.
- Does encryption prove compliance?
- No. Encryption is one technical control. Regulatory compliance also depends on governance, contracts, rights handling, processing locations and the facts of each use.