Voice
Decide whether members may send their voice through your organization's cloud, which models carry it and who picks them.
Feature Updated 19 Nov 2026Who can do this
Organizations on Enterprise Cloud. Owners and admins.
What the Voice tab decides
The Voice tab of Models holds your organization's voice models and the voice settings that go with them. Voice means dictation, where a member's speech becomes text, and spoken replies, where Panda reads an answer aloud. The tab decides whether members may send their voice through your organization's cloud and which models carry it. It also decides whether members may use their own OpenAI key.
Voice on the member's own computer is always available, because nothing leaves it. It never appears on this tab, because it cannot be restricted.
Where to find it
Open Models in the console, under AI, and pick the Voice tab. It opens at Voice. The tab lists your Voice models with Register model and Team access. Under them are three cards: your cloud for voice, the models members use, and members' own keys. To set a team's own values, choose the team in the menu at the top right. It reads Whole organization until you pick a team.
Your voice models
Voice models lists the speech to text and speech out models you registered, with the same switches, menu and Team access as chat models. Register one with Register model. It needs an OpenAI or Azure OpenAI provider key. See Models for the steps and Team model access for team access. Without a voice model the tab reads "No voice models registered".
Your cloud for voice
The first card is named after your cloud, for example Acme cloud for voice. It has two switches.
| Switch | What it does |
|---|---|
| Members can use Acme cloud | On, members may pick your cloud for voice in PandaOS. Off keeps every member's voice on their own computer. |
| Start members on it | Members who have not picked a voice option yet get your cloud. Nobody is switched without seeing it. Shown only while the first switch is on. |
A new organization starts with Members can use Acme cloud on and Start members on it off. The card also says: "Members who already used PandaOS Cloud voice move to your cloud automatically. Members on their computer are never moved."
Which model members use
The card Which model members use starts with Who picks.
| Choice | What members get |
|---|---|
| You pick one | Every member uses the model you choose. |
| Members choose | Members pick from every model their team can use. Yours is preselected. |
Under it are two rows, each with a menu of your enabled models of that kind.
| Row | What it is |
|---|---|
| Speech to text | How members' speech becomes text. Speech to text alone is enough for dictation. |
| Speech out | The voice Panda answers with. Optional, for spoken replies. A speech out model lists its voice after its name. |
With exactly one enabled model of a kind, the row uses it without a choice. With several, choose one, or members get no cloud voice for that part. Under Members choose, the rows read Preselected for members. Members start on the model you pick. If it is turned off or not granted, they start on the first ready model by name. Each menu ends with Register a model, which opens Register model on this tab.
Members' own API keys for voice
The card Members' own API keys for voice has one switch, Allow own OpenAI key. It starts off. On, a member may use an OpenAI key of their own for voice. Audio then goes to OpenAI on the member's own key, outside your usage and your provider accounts.
What a team can change
Choose a team in the menu at the top right, which reads Whole organization, to give it its own values. A team can only be stricter than the organization, so it never sees a choice it could not make.
- When the organization has switched its cloud off, the team's card reads "Your organization has switched cloud voice off, so it is off for this team too."
- Start members on it appears for a team only while the organization has it on.
- Under You pick one for the organization, the team reads "Your organization picks the model for every member, so this team does too." Under Members choose, a team may still pick You pick one.
- The model menus offer only models granted to the team. A team's page does not list the models, because they are the organization's. Its menus end with Register a model for the organization.
- When the organization does not allow own keys, the team reads "Your organization doesn't allow members' own keys for voice, so this team can't either."
A value the team sets itself carries a Custom badge and a reset icon labelled "Use the organization default". Without teams, the menu offers only Whole organization, and every member follows the organization's voice settings. See Set team overrides.
Publishing your edits
Edits on this tab go into the same Policies draft as the other policy pages. Nothing reaches members until you click Publish with the scope's name, such as Publish organization. Discard drops the edits. See History.
When a row shows a warning
A warning under a model row means members get no cloud voice for that part until you act.
The row says Choose a model
The full line reads "Choose a model. Until you do, members get no cloud voice for this part." More than one model of this kind is enabled and none is chosen. Pick one in the menu and publish.
The row says This model is no longer enabled
The full line reads "This model is no longer enabled, so members get no cloud voice for this part until you choose another". The chosen model was disabled or removed. Enable it again in Voice models, or choose another model.
The row says This team can't use this model
The full line reads "This team can't use this model, so its members get no cloud voice for this part". The chosen model is not granted to this team. Switch it on for the team in Team access, or choose a model the team has.
The row says Members can't dictate through your cloud
The full line reads "Members can't dictate through your cloud until a speech to text model is enabled." Your cloud for voice is on, but no speech to text model is enabled. Register or enable one in Voice models.