Skip to main content
Every call an agent makes is recorded under the person who asked. This page is the reference for roles, agent scope, skill visibility, API keys and the record. The settings are on Govern every agent. Sign-in is on SSO, and who may use, edit and share an agent on Permissions.

​
Where records live

​
Organization roles

Manage members and groups under Platform > Organization > Members & groups. Invite members by email. Groups can follow your identity provider’s groups through OIDC group sync.
Members & groups page listing members with role and status

Platform > Organization > Members & groups. Shown with sample data.

​
Agent scope

An agent reaches only the skills it was given. Admins decide which skills are available to the organization and can set any skill to Off for every agent in the Skills registry (Skills). A person adds available skills to an agent without approval. Skills an agent builds wait for an admin only when Publish without approval is off. Further limits:
  • Approval on a skill action: individual actions can require approval. A requester can withdraw a pending approval. Declining answers that one attempt, and the conversation continues.
  • Gated commands: hold matching shell commands on Claude Code, Codex or OpenCode.
  • Knowledge bases: attached per agent.
  • Budgets: cap what one agent may spend.
  • Who can send tasks: under the agent’s Channel settings, the Omni switch and Any agent, Selected agents or No agents. Each task shows which rule allowed it. See Invocation channels.
With a personal skill connection, the agent acts with that person’s own permissions, where systems allow. The outside system’s access control still applies, and its logs name the person.

​
Skill visibility

Each skill’s visibility decides which agents may hold it: The skill’s creator and admins (Root access or Admin) change visibility and allowlists. Others cannot widen a scope, and attaching a skill outside its allowlist is refused. The runtime builds each agent’s workspace from the skills its scope allows, so a scope change applies on the next run. Scope changes, publishes and archives are audited as UPDATE_SKILL_ACCESS, PUBLISH_SKILL and ARCHIVE_SKILL (see The record). Repository skills take their scope from xpander.yaml (Publish Skills). What the person adding a skill sees is on Skills.
The default is report in the current release, so teams can review violations first. enforce follows. On self-hosted installs the switch is XPANDER_SKILL_VISIBILITY_MODE (off, report, enforce). The air-gapped chart renders it from global.skills.visibilityMode, so set it in your values file.

​
Single sign-on

Sign-in with your identity provider uses OIDC, on every deployment. Setup, group sync and agent pre-authentication: SSO with your identity provider.

​
API keys

Air-gapped installs mint all credentials in-cluster at install time. Keep keys in your secrets manager and rotate them. Deleting an organization key revokes it within a minute at most.

​
The record

Per task: messages, each skill call with parameters and result, and each model call with cost, under the person who asked. In xpander Chat (Session Manager) and over the API (GET /v1/tasks/{id}/thread, GET /v1/tasks/{id}/llm_usage). Organization audit log: Platform > AI governance > Audit log, filterable by event type and exportable as CSV or JSON. See Audit and debugging.

​
Next steps

Set the controls

The settings behind each rule.

See an approval

The approver’s side.

Review security

Architecture per deployment.