How honest is your AI chat app about your data?

Ask an AI chat app what happens to your conversations and you will usually get one of two answers: a wall of legal text, or a reassuring word like "private" doing a lot of unexamined work. We think you deserve a third option—a plain answer. So here it is, for Ethyx.

For a product-agnostic plain-language walkthrough of where chat messages go and what to keep private, see AI 101 Part 5: What happens to what you type?.

What we actually protect

Your chat history is encrypted at rest. Messages, conversation titles and summaries, project notes, search queries, image prompts, and uploaded files are encrypted in our database. Encryption keys are stored separately from database storage, with access limited to what running the service requires. Connections to Ethyx use HTTPS.

Your identity is pseudonymized by default. Before a prompt reaches an AI provider, your name, email, and attachment filenames are replaced with reversible pseudonyms, then restored in the reply you see. If pseudonymization is ever switched off, Ethyx withholds your identity from providers entirely rather than sending it raw—it fails closed. For the full picture—operator API credentials vs your personal ChatGPT or Gemini account, how streaming restoration works, and a direct comparison table—see Your name never reaches the model.

We request no-retention where the provider API supports it. For OpenAI, xAI, and Gemini chat requests, Ethyx sends a per-request no-retention flag (store: false) by default. Anthropic and Perplexity state in their commercial API terms that they do not train on API conversations. Our model catalog shows a privacy rating for every model so you can see the posture of the provider behind it—including a clear caution label on providers that may store and train on your messages.

Your account can be locked down. Optional MFA with an authenticator app or email codes, recovery codes, and trusted devices are all in Security settings.

What we cannot protect—and won't pretend to

Providers receive your prompts. To answer you, the upstream model has to read your message. That text leaves Ethyx over HTTPS and is processed on the provider's infrastructure. Encryption at rest protects stored data on our side; it does nothing to the live request.

This is not end-to-end encryption. Our servers decrypt your history to assemble context for each turn. An AI client that never sees your plaintext cannot send it to a model.

A no-retention flag is a request, not a guarantee. We send it, and we show you where we send it, but we do not control provider infrastructure, and we do not have zero-data-retention agreements with providers. Anyone who tells you their multi-provider chat app guarantees your prompts are never retained is asking you to take a lot on faith.

Image contents are visible to the provider. We cannot pseudonymize what is inside a photo. The composer tells you this when you attach one.

Why we put the caveats in the marketing

The honest version of AI privacy is a list of trade-offs, and most marketing pages delete the second half of the list. We keep it in—on our features page, in the model catalog, in the composer itself—because trust built on vague wording does not survive contact with a data-handling question.

If you have a use case where these trade-offs are not acceptable—regulated health data, privileged legal material—the honest advice is that a cloud AI chat client may be the wrong tool, ours included.

For everything else: your history is encrypted, your identity stays out of prompts, and you can see each provider's posture before you pick a model. That is what we do. Now you know exactly where it ends.

---

Ethyx is in closed testing with an access code today. Subscriptions will launch after closed beta.