Skip to main content

Real-time voice

Real-time voice lets users have natural, hands-free voice conversations with Glean Assistant instead of typing. This page covers how to enable and roll out real-time voice for your organization. For end-user instructions, see the real-time voice user guide.

note

Real-time voice is available on the Glean Universal Model Key and Customer Key, with OpenAI or Azure OpenAI as the model provider. See Feature availability by LLM key for the full matrix.

Configure real-time voice​

Admins control real-time voice rollout from Admin console → Assistant → Real-time voice.

Three rollout options are available:

OptionDescription
OffReal-time voice is turned off for all users.
On only for adminsOnly admins can access real-time voice. This is the recommended starting point for evaluation.
On for everyoneAll users in your organization can access real-time voice.

Your selection can be changed at any time.

note

On the Glean Universal Model Key, real-time voice uses GPT Live 1 by default on version 561.0 and later, with OpenAI or Azure OpenAI. GPT Live 1 transcribes speech itself, so it doesn't use a paired transcription model. Earlier versions use GPT Realtime 2.1 with a paired transcription model.

note

On Customer Key, an admin must set real-time voice up separately. Choose a realtime model in the Model Hub first. Until one is configured, the setting stays off and can't be changed.

For the supported voice and transcription model pairs for each key type, see Feature availability by LLM key. On Azure OpenAI, the realtime model and any transcription model you select must be deployed in your Azure OpenAI resource.

You can also choose GPT Live 1 as your realtime model on Customer Key. It isn't selected by default. GPT Live 1 transcribes speech itself, so you don't pair it with a transcription model.

warning

Azure OpenAI is retiring GPT-4o Transcribe on October 15, 2026. If you use Azure OpenAI, switch to GPT Realtime 2.1 with GPT Transcribe before that date. GPT Transcribe needs version 559.0 or later.

  1. Start with admin-only access: Set the feature to On only for admins to evaluate the experience.
  2. Test use cases: Have admins test real-time voice in various scenarios to understand value and behavior.
  3. Expand gradually: Once comfortable, enable for all users or specific groups.
  4. Monitor feedback: Collect user feedback through standard channels to inform adjustments.

Privacy and data handling​

  • No raw audio is stored. Audio streams are routed from the user's browser to your configured model provider (OpenAI or Azure OpenAI) and back. No audio recordings are retained.
  • Transcriptions are stored like text chats. Voice transcriptions follow the same storage and retention policies as standard chat messages.
  • Permissions are enforced. Glean's existing security model and document-level permissions apply to all voice interactions, the same as text-based chats.

Usage and pricing​

Real-time voice usage may be subject to usage-based pricing.

Limitations​

  • Real-time voice currently works with the general Glean Assistant. It doesn't automatically route voice requests to published agents, and users can't start a voice session with a specific published agent. To use an agent or automatic agent routing, users must switch to Chat. For more information, see Route requests to agents.
  • Only document artifacts can be created from voice at this time. Slides, HTML, and other visual artifact types must be created from text chat.

See also​

Last updated on
© 2026, Glean Technologies, Inc.
Was this page helpful?
On this page