Before you begin
- Organization Admin role. Most steps need it. See Organization roles.
- A second administrator. Plan for one, so access never depends on one person.
Checklist
1
Decide how people sign in
Verify your company email domains, choose how people with those addresses join, and connect your identity provider.
2
Create team workspaces
A workspace decides where skills and agents live and who can see or change them. Create one for each team that shares skills, then choose whether its changes release automatically or wait for approval.
3
Invite members
Invite people to the organization, or straight into the workspaces they need, with the right roles. Give a second person the organization Admin role.
4
Roll out Atlan Desktop and the trace plugin
Atlan Desktop brings each person’s Claude Code and Codex skills and sessions into AI Registry. On managed computers, deploy it with your MDM. To capture Claude Code sessions across a fleet, deploy the trace plugin with managed settings.
5
Choose what is stored from sessions
Decide whether Atlan stores session content, or only the run’s metadata. Settle this before people start sending traces.
6
Connect models and set rates
Connect the model providers your agents should reach through Model Gateway. Then set the rates your organization pays, so run cost matches your bill.
7
Install optional apps
Metrics Studio and Ontology are installed by an administrator. Install them only if your teams will use them.
8
Create credentials for automation
Scripts, CI jobs, and services use API keys instead of a person’s sign-in. Scope each key to what it needs.
Verify
- A colleague you invited can sign in, sees the team workspaces they joined, and finds a shared skill. If they can’t, see Workspace access errors.
- Settings shows your verified domains and your second administrator.
Next steps
- Admin & security: every administration task in one place.
- Availability and limits: what is on by default and what needs turning on.