--- title: Agent keys description: Each coding agent you connect gets a key that you can name, narrow, limit in time and revoke. --- # Agent keys When you connect a coding agent (Claude Code, Cursor, the Kaviris CLI on another machine), it gets an **agent key**: your access, for that agent, in one organisation. You manage them in the portal under **Agent keys**. ## What a key may do A key can never do more than you can. What it may do is: **your rights right now ∩ the key's permissions ∩ the key's apps and groups** - **Your rights right now.** If you lose a right, or leave the organisation, every key of yours loses it at the agent's next call. - **The key's permissions.** Pick a preset or choose single permissions: - **read**: see apps, logs and the organisation. - **deploy** (the default for a new key): read, plus create and deploy apps, edit and approve contracts, set secrets, roll back, take backups. No deleting, restoring, exporting or administration. - **custom**: any set from the list. Permissions marked *needs approval* are only ever asked for; a person [approves](approvals.md) them. - **Apps and groups.** Leave "all apps", or limit the key to some apps and groups. A limited key never acts on the organisation itself. ## Managing keys For each key you see its name, the agent (client), when it was created and last used, its permissions, its apps and its expiry. You can: - **rename** it, for example "Claude Code – laptop", so audit entries and approval requests say which machine asked; - **change its permissions** or its apps and groups; - **set an expiry date**, or none; - **revoke** it. The agent's next call is refused; connect it again to give it access back. Only you edit your own keys, in the portal; an agent cannot change its own key. Owners and admins see every key in the organisation and can revoke any of them. Every change is recorded.