Choose the model.
The team decides whether to use Codex, Claude, or a supported local model.
Product data handling
Clear boundaries build useful trust.
This page explains the current Varta 0.3 architecture in plain language. It is a technical product summary, not a legal privacy policy.
Last updated September 20, 2026
On your machine
The room starts local.
| Transcription | Microphone input is transcribed on the machine. |
|---|---|
| Speaker recognition | Speaker identification is handled locally. |
| Meeting history | Conversation history is stored locally by the application. |
| Raw recordings | Raw microphone recordings are not saved as the meeting history. |
Configured services
Cloud use is explicit, not invisible.
If you configure a cloud model or cloud speech, some information must leave the device for those features to work.
| Model prompts | The current prompt and relevant conversational material are sent to the model provider you configure. |
|---|---|
| Project context | Only project context selected for the room is included when the model needs it. |
| Speech generation | If optional ElevenLabs speech is enabled, Varta sends the answer text to ElevenLabs to generate spoken audio. |
| Provider policies | Your configured provider's own retention, security, and privacy terms apply to information sent to that provider. |
Your control
Know the boundary before the room begins.
Choose the context.
Project material is selected deliberately rather than gathered indiscriminately.
Choose the voice path.
Cloud speech is optional. The room can operate without sending answer text to a speech provider.
Before using Varta with sensitive information, review the terms and data controls of every external provider you configure.
The principle
Privacy claims should be precise enough to act on.
We will keep this summary aligned with the product as Varta moves toward public release.
Explore Varta