Skip to content
SOVEREIGN

Legal

Privacy Policy

Last updated: 2026-09-26

1.What data we collect

  • Account: email, name (optional), authentication identifier.
  • Usage: number of requests, selected model, plan status — to provide the service.
  • Chats and memory: your chats, the facts saved by memory and your knowledge-base documents are stored in our database (Supabase), linked to your account. Only your account can read them and the database provider encrypts them at rest, but they are not end-to-end encrypted — our server reads them in order to answer you.
  • Payment: processed by the payment gateway (Dodo Payments / crypto / RollyPay SBP); we do not store your card details.

2.Your chats and AI models

  • Where they go: each message is sent from our server to the AI provider that runs the model you chose. Depending on the model and availability: OpenRouter, OmniRoute, Perplexity, Groq, Cerebras, SambaNova, Mistral, OpenAI, NVIDIA NIM, RSI AI. OmniRoute is our routing gateway; it forwards the request to the providers connected to it. If the main providers fail, free backup gateways are used (LLM7, Experiential Labs). Text embeddings for the knowledge base and answer cache are computed via OpenRouter.
  • Which model answered: under every reply we show the model that actually answered and the tokens counted for that reply. If the model you chose failed and a different one answered, we say so openly ("requested: X → answered by: Y").
  • Provider policies: these providers process requests under their own terms and privacy policies; some may store or log requests. We do not control this and cannot guarantee that they do not use data for training. For sensitive data, turn on private mode (Blind Prompting).
  • Training: SOVEREIGN does not sell your chats or give them to providers for training. The only exception is our own model, Tella: if "Train Tella 2" is on in Settings (it is on by default), short question-answer pairs from ordinary chats may be saved without your account ID. Never saved: chats that used files, the knowledge base, memory, connected services, links you sent, your own skills or private mode, and any text where an email, phone number, card number or API key was detected. Turn it off in Settings → Data & privacy and no new samples are collected.
  • Answer cache: simple questions without personal context (no memory, files, links or knowledge base) are written to a shared cache as question and answer text, without your account ID, and are used for 24 hours to answer identical questions faster — including for other users.
  • Generated images: generated images and videos are kept only in your browser; only a placeholder is synced to our server (and your other devices) in their place.
  • Deletion: you can delete a single chat from the list; Settings → Data & privacy → "Delete all data" deletes all your chats (with their messages), memory, shared links, knowledge-base documents and connected services from our database, and signs out your CLI and Crownwork devices. Training samples and cache entries are stored without an account ID, so they cannot be traced back to you and are not removed this way. To delete your account completely, email us.
  • Crownwork Mods: when you're signed in to Crownwork and create, change, enable/disable or suggest a mod, its name, type and content (theme colors, command text or panel code) are stored linked to your account and visible to our team — the Mods window says so too. Your files, chats and folder paths are never sent; only if you enable your own interface in Crownwork Studio are the interface files you changed sent too. Approved suggestions are published in the gallery for everyone without naming the author. These records are deleted with your account; email us to delete them sooner.
  • Sovereign Redact (Crownwork): before sending a request to the server, Crownwork replaces API keys, passwords, emails, phone, card and ID numbers (and words you add) with masks like [KEY_1], [EMAIL_1] on your computer; the real values never reach the server or the models and are put back into the answer on your computer. This is pattern-based and not a guarantee: text that matches no pattern (such as a name or an address) is sent as usual. It can be turned off in Settings → Privacy.
  • Mistake book: when an answer is marked wrong (by you or by the independent judge), that question, answer and the “what was wrong” note are stored linked to your account and a short rule is derived from them. Rules are added to your following requests only after you approve them; you can edit or delete them at any time (Settings → Mistake book). These records are never used for advertising or sold; for model training they may be used only with separate, explicit consent and only as anonymous rule text (never the question/answer).

3.Blind Prompting

When private mode (Blind Prompting) is on, sensitive fragments such as names, numbers and company names are masked in your browser before the request is sent to an AI model, so AI providers do not see them. Masking is automatic and pattern-based and may miss some details. When the mode is off, the request text is sent to the provider unchanged.

4.How we use data

We use data only to provide the service, manage your account, ensure security and meet legal obligations. We do not sell your data and do not share it with third parties for advertising.

5.Third-party services

To operate the Service we use: AI model providers (to process your requests — listed above), hosting (Vercel), a database (Supabase) and payment gateways (Dodo Payments, crypto). Each has its own privacy policy.

6.Retention period

Your data is kept for as long as your account is active. You can delete your account or individual records at any time — once deleted, they are removed from the system.

7.Your rights (GDPR)

You have the right to request access to, correction, export and deletion of your data. Your memory (saved facts) can be exported in full and deleted at any time.

8.Cookies and local storage

Only essential cookies (session, authentication) and the browser's local storage (preferences) are used. There are no tracking or advertising cookies.

9.Children

The Service is intended for users over 18. We do not knowingly collect data from children.

10.Contact

Privacy questions: isa704480@gmail.com. More details: Terms of Use.