Join the Waitlist

How Avori uses AI providers

Updated November 3, 2026 (version 1). This is what you agree to when you tick “I agree Avori sends my messages, files and connected-app content to AI providers like Anthropic, OpenAI, Google and xAI to do the work I ask.”

Who receives your data

Avori’s agents are powered by AI models from these providers. Depending on the model your agent uses, your data can go to:

  • Anthropic (Claude)
  • OpenAI (ChatGPT, Codex)
  • Google (Gemini)
  • Meta (Llama)
  • Moonshot AI (Kimi)

This applies however the model is paid for: with Avori credits, your company’s own API key, your personal ChatGPT or Claude plan in Avori’s cloud, or the apps on your Mac. If Avori adds a provider that isn’t on this list, we ask you again before anything is sent to it.

Voice typing is transcribed by Groq (OpenAI as backup).

Voice features

Voice typing sends your recording for transcription; a supported fallback can send the same recording to OpenAI. You can review the text before sending it. Live calls use OpenAI and can save transcripts in your conversation. Voice notes you attach are saved files. Provider retention and settings depend on the route and account described in the privacy policy.

What is sent

  • Your messages and what your agents write back.
  • Files you share with your agents.
  • Content from apps you connect (for example email, documents or store data), when your agents use them for a task.
  • Your voice, when you use voice typing.

Why

To generate your agents’ responses and do the work you ask, including routines and triggers you set up.

How it’s handled

Avori sends only what a task needs. With Avori credits or your company’s API key, the provider’s business terms apply. With your own ChatGPT or Claude plan, the provider’s consumer terms and your plan’s settings apply (or its business terms, for a work plan). Avori doesn’t sell this data or use it for advertising. Your private chats stay private to you inside Avori unless you share them.

Your choices

You can turn this off at any time in Settings › Privacy. Revoking permission blocks new model requests through the agent, routine and trigger routes that use this permission. It does not recall data already sent. Voice typing and live calls have separate processing, described above. Stopping a dictation recording submits it for transcription; ending a call stops the live session. Neither recalls data already sent. Your settings, your account and deleting your account keep working either way. Each person decides for themselves; an agent or an administrator can’t agree for you.

For the full details, read our privacy policy.