privacy

Privacy isn't a feature here. It's the foundation. Your messages to Maya, Sage, and Luna are encrypted at rest with a key tied to your account, and even people who run talkamore, me included, can't read them by default. The product is built so that the only person who can see what you wrote is you.

Your messages, any journal entries we've rolled them into, the wiki pages built from your own words, your display name and timezone, and operational metadata like when you last sent a message. That's it. We don't read anything outside this product and we don't link you to anything outside it.

Every message body, every journal entry, the few-line context you typed during onboarding, your display name: all encrypted with a key that's unique to your account. The encryption happens before anything is written to disk. If someone got hold of a database backup tomorrow, they'd see opaque bytes where your conversation used to be.

The keys themselves are wrapped with a master key that lives in our deployment environment, never in the database. So a database leak alone doesn't leak your data, and a server leak alone doesn't either.

By default, nobody on the talkamore side can read your message text. The admin dashboard the operator uses shows counts, dates, mood scores, and message lengths, never the words themselves. Profile fields that contain free text (your "what's going on lately", your saved preferences) are surfaced as lengths only. The product is wired so the operator path literally never holds a decryption key in scope.

There's exactly one exception: if Maya flags a turn as a safety concern (someone in active crisis), the operator can read that note to triage. That single page is being moved behind a break-glass step-up auth that records a justification on every access. That follow-up is already in flight.

Every admin view is logged: who looked at whose account, when, from which IP. That log can't be edited from the dashboard. We don't sell data. We don't share with advertisers. There is no third-party analytics of your messages.

Encrypted message history, journal entries, and your wiki pages all sit in a Postgres database on Railway, every row scoped to your user id so your data never mixes with anyone else's. Long-term recall runs on that same wiki, in-house, with the same encryption-at-rest as everything else. There is no external memory service holding your content.

Model inference happens on cloud model APIs (currently AWS Bedrock). Your messages are sent to the provider's servers to generate a reply, then we throw the raw response away and keep only the text the assistant said back. The provider's API data-usage policies apply: they don't train on your messages.

If you tell Maya something that looks like active crisis (a stated plan to harm yourself or someone else), we record a small flag with a one-line summary and a severity level. For HIGH severity, Ishan gets a notification so a human knows it happened. This exists so we don't ignore you in a moment that matters. It is not shared with anyone outside talkamore.

Send Maya the message "delete my account" (or equivalent, "wipe everything," "remove me," she'll ask you to confirm). On confirmation, every database row keyed to your user id is deleted, wiki and stored files included. The deletion is immediate and cascades across all tables.

If you prefer, email team@talkamore.com and we'll do it manually within 48 hours.

If you bought lifetime access, the transaction runs through Dodo Payments. We store the transaction id, amount paid, and timestamp, no card number, no billing address.

We won't email you marketing. We won't add you to a newsletter. We won't show your messages to anyone outside a specific bug investigation. We won't train a new model on your data.

Questions, concerns, or a deletion request: email team@talkamore.com. A real person reads it.