AI access
Choose which AI members can use: which of their own AI accounts they may connect, which models they may pick, and where models may run.
Feature Updated 19 Nov 2026What AI access is
AI access decides which AI members can work with in PandaOS: the AI your organization provides, and the AI accounts members bring themselves. It also limits the models on each account and where models may run. Set it before members start, because every account a member brings stays off until you publish.
Where to find it
In the console, open AI access, the first page under Policies in the sidebar. The Policy scope menu at the top switches between Whole organization and each team. Owners and admins can change it.
Enterprise Cloud is always on
On organizations with Enterprise Cloud, Enterprise Cloud reads Always on under "Provided by your organization". It runs the models you add under Models, paid with your provider keys, and members need no setup. A policy cannot turn it off. Which models a team gets is set on the Models page, see Team model access.
Allow members' own accounts
- Open AI access.
- In the Policy scope menu, choose Whole organization or a team.
- Under "Members' own accounts", choose On, Off or Blocked for each account.
On and Off set the default every team starts from. Blocked turns the account off for the whole organization, and no team can turn it back on. Teams choose only On or Off. When teams turned an account on for themselves, blocking keeps their setting stored with no effect, and unblocking brings it back into effect. - Click Publish organization, or Publish with the team's name, such as Publish Sales.
Accounts you can allow
| Account | What it is |
|---|---|
| Claude subscription | The member's own Claude plan |
| Anthropic API key | A personal Anthropic API key |
| AWS Bedrock | The member's own Bedrock account |
| Codex | The member's ChatGPT or OpenAI account |
| OpenCode Go | The member's OpenCode Go subscription |
| Personal API keys | OpenAI, OpenRouter or Azure keys a member adds |
| Models on their computer | Ollama and other local runners |
| Custom endpoints | OpenAI-compatible URLs a member adds |
| Free models | PandaOS's shared free pool |
All of them are off until you publish.
Limit the models on an account
- On the account's row, click ..., then Configure models.
- At the top of the dialog, choose Select.
On a team, Follow organization keeps the organization's list, including its later changes. - Pick models from the list, or enter one under Add exact model ID and click Add ID.
Claude subscription and Anthropic API key offer a model list. For the other accounts, enter the exact model IDs that account uses. - Close the dialog and publish the scope.
With no model selected, publishing blocks every model on that account.
A team list replaces the organization list
A team's model list for an account replaces the organization's list for that account, and can allow models the organization's list does not. No model list can be locked for teams. A model rule never turns an account on: the account must be allowed as well.
Restrict where models run
- On AI access, choose Whole organization in the Policy scope menu.
Allowed hosting applies to every team. Teams cannot change it. - Under Allowed hosting, tick the regions members may use.
The choices are Global, EU hosted, Germany hosted, and any other tag already on your models. - Click Publish organization.
Tag your models first
Once a region is ticked, models without a region tag stop being offered. Tag your models on the Models page before you publish. With nothing ticked, hosting is not restricted.
What members see
- An account that is off or blocked disappears from the model picker and from their connection settings. A new chat on it opens on Enterprise Cloud instead.
- Model pickers list only the allowed models.
- Enterprise Cloud models outside the ticked regions are not offered. Settings, Connections, Hosting reads "Your organization allows only EU hosted." Members can narrow this in their own settings, never widen it.
Members get the change within about 5 minutes, and sooner when they restart PandaOS, wake their computer or click Refresh in Settings, Organization.