AI and your data
This page explains how Omne's AI features and agents use your data: which AI providers are involved, what they receive, and what stays in your control.
Which features use AI
- Ask AI, the chat in your workspace, which can read and, with your yes, change things through Omne's tools.
- Questions in link analytics, where a sentence you type becomes filters for your clicks.
- Your own agents, such as Claude Code or Codex, connected from your computer or run in Omne's cloud.
- Voice, which writes down your voice notes to Omne AI, your dictation on the web and your podcast episodes, and reads answers aloud.
Which providers
Every AI feature but voice and Smart's choice of model runs on an AI provider you connect yourself, in Settings under AI providers: with your own key, or by signing in to your own plan. OpenRouter connects in one click; OpenAI, Anthropic, Google Gemini, xAI, Mistral, DeepSeek, Alibaba's Qwen and others take a key. That provider processes your requests under your own account and its terms, and bills you there. Omne keeps the key sealed and never shows it again.
Voice runs on Omne's own account, on Omne's own speech server in Germany. Every language is heard there: none of your voice goes to anyone else. Smart, which picks the model for each message in Omne AI, runs on Omne's own account too: TypeSafe, in the United States, reads the message and the chat's last two turns, never names or connected accounts, only to choose which model answers. TypeSafe doesn't train on them, and is on our Subprocessors page too. Pick a model yourself and Smart isn't asked. Other AI included with an Omne plan, on Omne's own key, is not live yet. When it is, its provider will be on our Subprocessors page before it receives any data.
What is sent
When you ask Ask AI something, Omne sends the provider your message, the chat so far, a short description of your workspace (its name and the accounts connected to it), and what Omne's tools return while it answers, such as the posts, notes or figures that answer your question. A question in link analytics sends only the sentence you typed and today's date.
No training on your content
Omne does not train AI models on your content, and does not sell it. The provider you connect handles your requests under your own account, so its settings, such as whether it keeps what you send, are yours to choose there.
What agents can do
An agent works only through the tools and accounts in your workspace, and only during a turn you started: its tools answer to a key that exists for that turn alone. An agent in Omne's cloud gets its own sealed computer at Daytona, in the EU, running the maker's own program signed in to your own plan. That sign-in happens on the maker's own screens and stays on that computer; Omne passes your one-time code straight in and keeps nothing of it.
Approval before any change
An agent can read what you let it read. Anything that changes something, such as publishing, sending, editing or deleting, waits for your yes, and you see what it wants to do first. With no answer within 10 minutes, the answer is no.
The audit log
Changes an agent makes to Finance are recorded with the agent's name. Changes made in Omne's admin area and events on connected accounts are recorded too.
How to revoke access
Remove an AI provider in Settings, under AI providers, and Omne deletes its key. Remove an agent, and Omne deletes its key; for an agent in Omne's cloud, its computer and the sign-in on it go too.
AI can be wrong
What AI produces can be wrong, incomplete or similar to what others receive, so check it before you rely on it or publish it. Once Finance is live, its tax and money drafts are checked before they are used: by you, and by a qualified person before any filing is sent.
How AI-made content is labelled
Omne does not add a label to what you make with AI. When you post to TikTok, you can mark a post as AI-generated, and TikTok shows its own label. Where a platform or the law asks for AI-made content to be disclosed, that is up to you, as our Terms of Service say.