Skip to content
chat

Privacy

The behaviour described here is the behaviour the code implements today.

Everything here is public

Kevin has one public global room, one public room per profile, and public profiles. There are no private rooms and no private messages. Anything you post or save is public.

What Kevin stores

  • Account. A pseudonymous account identifier. Kevin never asks for an email address, phone number, password or social login.
  • OAuth hashes. Authorization codes and tokens are stored only as salted one-way hashes. The identifier your assistant sends (openai/subject) is stored only as a one-way HMAC digest, never raw, and is used only to recognise a returning account and to apply rate limits. It never authorizes anything by itself.
  • Profile data. Handle, name, bio, language, up to 10 links and up to 50 structured entries. Kept until you delete them or delete your account.
  • Room messages. Original text only, at most 500 characters. Kevin never stores translations. Messages, their mentions and their conversation assignments are physically deleted 30 minutes after posting by a scheduled cleanup job.
  • Activity. Compact activity records (mentions, follows, likes, new entries) with no message text. Room-linked activity expires with its message; profile activity is deleted after at most 30 days.
  • Profile views. Anonymous aggregate counters only — a total and a last-seen counter. Kevin stores no visitor identity, no IP address, no user agent, no timestamp and no visit history.
  • Operational metadata. Rate-limit buckets are keyed by a bounded HMAC digest — never a raw account identifier, subject, IP address or user agent. Idempotency records store only a short status reference, never message, profile or entry text, and expire within 30 minutes for room content.

Processing by AI models

Before any public text is stored, Kevin sends it to OpenAI’s free Moderation endpoint together with local structural checks. If moderation is unavailable, Kevin fails closed and refuses the write. Rejected text is never stored. Kevin stores no translations and creates no embeddings of your data. Private suggested-people matching involves no backend model call: Kevin only prepares candidates deterministically from already published profile and entry text, and ChatGPT itself judges who fits. One further model feature is switched off by default and can only be enabled by an operator: an optional News observer, which summarises short excerpts of already public room text. The public News page itself is produced deterministically, without any model call.

What Kevin does not do

No behavioural advertising profiles, no sale or sharing of data, no classic behavioural analytics, no private rooms, no user-to-user payments and no background push notifications. Kevin carries no advertising and no paid placement at all, and it sells nothing inside a connected AI assistant. Kevin is text-only: it stores no images, profile pictures or uploaded files at all. If you have an optional website subscription, payment runs entirely through the external payment provider on kevinchat.com, never inside a chat: Kevin stores only the resulting subscription state and never processes or stores card details.

Deletion

You can delete your whole Kevin account from ChatGPT at any time with the delete_my_kevin_account tool. It permanently removes your profile, entries, links, messages, social connections and activity. See the data deletion page for details.