PandaOSPandaOS
Enterprise

How policies work

Know what a policy controls, which console page sets each part, and where to start.

Overview Updated 19 Nov 2026

What policies are for

A policy decides what members may use in PandaOS and what the agent may do for them. You set it once for the whole organization in the console, and change it for a team where that team needs something else. Members get it in the app with no setup.

The organization policy sets the value every team starts from. A team can set its own value for a setting, and that value replaces the organization's for its members. Each member gets the combination for their own team, and the panel What Sales members get shows that result for a team.

Pages in this group

The console's sidebar holds the policy pages under Policies. The Policy scope menu at the top of every one of them except History switches between Whole organization and each team.

PageWhat it coversConsole page
AI accessMembers' own AI accounts, the models on each, and where models may runAI access
Model tiersWhich Enterprise Cloud model each tier usesModel tiers
PermissionsWhat the agent may do without asking, and which chat modes members getPermissions
Apps and MCPWhich apps and MCP servers members can connect, and rules for single actionsApps and MCP, with the tabs Connected apps and MCP servers
LibraryWhat members can share with colleagues, and what waits for reviewSharing
PluginsWhich plugins from the directory members can addPlugins
Set team overridesOne team's own settingsAny page above, with a team chosen in Policy scope
HistorySending edits to members, throwing them away, and bringing back an earlier revisionHistory
Policy settings referenceEvery setting with its options and default

Start here

Open AI access first, because it decides which AI members can work with at all. Then publish, so your edits reach members, see History.

Settings locked for teams

On the organization scope, the lock next to a setting locks it for teams. A locked setting can only forbid. It never forces something on.

Kind of settingA team's own valueWhen the organization locks it for teams
On or off switches, such as custom serversReplaces the organization's valueSwitch it off first. No team can turn it on.
Lists of allowed items, such as appsReplaces the organization's listChoose Select first. Teams can only narrow the list.
Permission levelsReplaces the organization's levelTeams can be stricter, never more permissive.
Members' own AI accountsOn or Off replaces the organization'sChoose Blocked. No team can turn that account back on.
Action rules for apps and MCP serversReplaces the organization's unlocked rulesEvery team gets the rule and cannot remove it.
Allowed custom server patternsReplaces the organization's patternsThe organization's patterns replace every team's.
Allowed hostingTeams cannot set itAlways locked for teams.
Model tiersReplaces the organization's pinCannot be locked.

What applies before you publish anything

Until the first publish, every team gets the starting policy:

  • Enterprise Cloud is on. Members' own AI accounts and custom endpoints are off.
  • All PandaOS apps and all catalog MCP servers are allowed. Custom servers are off.
  • File writes, shell commands and the browser are on Ask. Apps and integrations are on Ask for changes.
  • The All permissions and Auto-edit chat modes are allowed. There are no action rules.
  • Models may run in any hosting region, and PandaOS picks the model for every tier.
  • Remote machines are off.

The publish bar reads "Nothing published yet. Every team gets the baseline until you publish."

Drafts and revisions

Your edits stay in your browser tab until you publish. There is no saved draft: publishing is saving. Members never see an edit you have not published.

Edits are kept per scope while you move between the pages under Policies, the Voice and Image tabs of Models, and between scopes. Leaving for History drops them, so the console asks first.

Every publish becomes a numbered revision. The numbers are one counter for the whole organization, so publishing a team also moves the revision number. A published revision is never changed. To undo, you restore an earlier one, which publishes it again as a new revision.

When members get a change

PandaOS checks for a new policy every few minutes. It also checks when it starts, when a member signs in and when the computer wakes. The same happens when the member switches back to PandaOS and when they click Refresh in Settings, Organization. A publish reaches members within about 5 minutes, and often sooner.

The change applies to open chats too. A newly disallowed MCP server is removed from open chats. A chat in a chat mode that is no longer allowed restarts in an allowed one. An idle chat on an AI account that is no longer allowed moves to Enterprise Cloud.

When PandaOS cannot reach your organization

PandaOS keeps the last policy it received and uses it for up to 24 hours after the last successful check. Settings, Organization then reads "Using the last known policy. PandaOS checks again on its own."

After 24 hours, or when the computer holds no earlier policy, PandaOS locks down until it reaches your organization again. When access stops lists what members lose and keep, and the messages they see.

Settings that need a current PandaOS

Three settings need a current version of PandaOS on members' computers. They are switching off a chat mode, adding action rules for apps or MCP servers, and setting allowed custom server patterns.

Once you publish any of them, an older version cannot read the policy and locks down, as When access stops describes. Its members see "Your organization uses a newer policy than this version of PandaOS reads. Update PandaOS to continue." Ask members to update before you publish these settings.