Open WebUI 0.12.0 adds real-time voice calls, two-factor sign-in and shared chats

Open WebUI’s 0.12.0 release adds real-time voice calls, two-factor sign-in and a way to collaborate in shared chats. It is a substantial step from chat interface towards AI workspace, with new controls for administrators and users as well as a few sensible brakes on the machinery.

Open WebUI Watch analysis

What happened

The project’s 0.12.0 release notes say the version was published on 10 October. The changes span voice, account security, collaboration and workspace management. Here are the ones with the clearest practical payoff:

Our top picks

• Real-time voice calls
  Voice mode can use OpenAI’s gpt-realtime-2.1-mini by default, handing questions and tasks to the chat’s selected model.
• Animated voice-call avatars
  Models can use uploaded VRM characters, optional animation clips and up to 16 named gestures.
• Two-factor sign-in
  Administrators can require authenticator-app codes; users receive ten single-use recovery codes.
• Replies in shared chats
  Owners can let people or groups continue the same conversation instead of making separate copies.
• Controls for model settings
  Model editors can offer named options, such as low, medium or high reasoning effort, for users with the right permissions.
• Compaction during long tool runs
  With context compaction enabled, long sequences of tool calls can be summarised between rounds rather than only before a reply begins.
• Skills with supporting files and version history
  Workspace skills can include scripts, references and templates, with saved versions users can compare or restore.
• A file-browser view for knowledge bases
  Users can browse files, inspect the text the model searches, edit it and download the original. The release notes also say real-time calls require the Allow Call permission and end after an hour. Tool approvals and questions still need answers in the chat, so the voice feature does not quietly become an unattended agent with a microphone.

Why it matters

These additions make Open WebUI more capable as a shared place to work with models. Real-time voice brings a quicker conversational route into a chat’s existing model and tools; shared replies could make that chat useful to a group rather than a set of parallel copies. For administrators, configurable two-factor authentication is a concrete account-security option.

The new controls and file tools also offer more ways to shape what models can do and what material they can use. That flexibility is useful, but it makes permissions and configuration more consequential. The release notes describe the available features, not how each will behave in every deployment.

Our read

There is enough here to make 0.12.0 a meaningful upgrade, not a version number with a fresh coat of paint. The most immediately useful changes are shared-chat replies, authenticator-based sign-in and the richer handling of skills and knowledge files. Voice calls are the showier addition, with an important distinction: the real-time voice model handles small talk, while the selected chat model handles substantive questions and tasks.

If you run Open WebUI, check the release notes against your setup and permissions before enabling new features. Administrators will want to decide who can start calls and whether to require two-factor sign-in; users may find the collaboration and file-management changes more immediately useful than the animated avatar, however charming its gestures.

What to watch

• How administrators configure call permissions and two-factor sign-in in real deployments.
• Whether shared-chat replies work well for teams using different roles and permissions.
• How real-time voice behaves when the selected model uses tools or needs user approval.
• Which release notes clarify compatibility and configuration changes for existing installations.

Discussion spark: Would you enable real-time voice and shared replies in a self-hosted AI workspace, or do the extra permissions and collaboration controls make you prefer keeping chats strictly personal?

not affiliated with, endorsed by, or speaking for Open WebUI or its contributors