Most fleets start the same way: a public key in authorized_keys on every host, and perhaps a bastion in front. It works until someone leaves, a laptop is lost, or an auditor asks who could reach production last March.
The problem with static keys
- Keys outlive their reason. A key added for one deployment is still trusted years later. Nothing expires unless someone remembers to remove it.
- Offboarding is a hunt. Removing a person means finding their key on every host, in every account. Miss one and they still have access.
- Access is invisible.
authorized_keystells you which keys work, not who holds them, why they were added, or whether anyone approved it. - Production is just another host. Nothing in the mechanism distinguishes a demo box from the primary database.
What Shellius changes
| Static keys | Shellius | |
|---|---|---|
| What a host trusts | Each individual public key | One certificate authority |
| Lifetime of access | Until someone deletes the key | The certificate's validity window, set by policy |
| Removing a person | Edit every host | Revoke once; the next connection is refused |
| Production | Same as everything else | Goes to an approver, unless a role may bypass — and that is audited |
| Record of access | Whatever sshd logged locally | Central, append-only audit log, plus SSH session recordings |
| Adding a host | Distribute everyone's keys to it | One bootstrap command |
What it does not change
Shellius is honest about its edges:
- It is a new critical service. Hosts ask it about every login. If Shellius is down, certificate logins stop. Keep a break-glass path that does not depend on it.
- It does not replace your network. Shellius has no tunnelling. Something still has to be able to reach port 22 — either the Shellius host (browser sessions) or the user's machine (terminal client). See Network reachability.
- Some hosts cannot be bootstrapped. Appliances and machines you do not control cannot trust a CA. The Keystore exists for those, as a deliberate and audited exception.