Not effective · Practices under review

Privacy · Draft for review

This is a non-effective draft, not a finalized Privacy Notice. Do not share secrets or highly sensitive information in the beta.

Private by design, not end-to-end encrypted

The server processes account identifiers and stored conversations in order to provide private history and lexical search. Full-text indexes can contain sensitive content. Conventional server processing is not end-to-end encryption.

Browser storage

Only validated palette, light/dark/system and motion preferences are saved in localStorage. Drafts, messages, custom personality prompts, model selections, balances and login tokens are not saved there. Private content is held in memory and cleared on logout. There is no offline transcript cache.

Operator access

The intended policy is no routine operator review of chat content without explicit, scoped permission. This is a policy, not a technical inability to read server-held data. Do not send transcripts or login links to support.

Conditional beta processing

Where enabled by the operator, text inference uses the processors disclosed in the private-beta notice. Check your signed-in account for current feature availability. Billing and provider web research are disabled in this beta. Image uploads, where enabled for your account, require separate processing consent. Enabled attachments use the image model disclosed in the in-app image policy, not necessarily your selected text model. No blanket zero-retention claim is made.

Approved inference processors

When inference is enabled, bounded conversation context is sent to Venice AI using the selected conversation model: Skinny (GLM 4.7 Flash Heretic) or Average Joe (Venice Uncensored 1.2), when that model is enabled by the operator. The owner selected an unmoderated, invite-only, non-paying private beta: OpenAI moderation is postponed, and no beta text is sent to OpenAI in that mode. There is no independent verification of the no-erotic-chat boundary. A later moderated mode would separately send input/context and the complete reply to OpenAI’s moderation API, hold replies until inspection passes, and require updated acknowledgement. Venice processes text under its applicable terms and data practices; this is not end-to-end encryption or a blanket zero-retention promise. Output can be inappropriate or inaccurate. Owner approval does not certify enforcement or provider endorsement.

Retention and your requests

Final retention, physical cleanup, backup handling and data-rights schedules are not approved in this draft. Conversation deletion is available after login; it must not be interpreted as immediate forensic erasure from all backups. Contact the beta operator for privacy or account requests.