Skip to content

Workspaces

A Rookery installation has one owner — the person who installed it — and any number of workspaces.

A workspace is a sealed world. It has its own:

  • knowledge base
  • stored credentials
  • connected accounts
  • agents and their schedules
  • choice of model

Nothing crosses between them. An agent in one workspace cannot read another’s notes, use another’s connected accounts, or see another’s credentials.

The common reason is keeping things genuinely apart — work and personal, or one per client. Because credentials and connected accounts are per workspace, an agent you build for one cannot reach into the other even by mistake.

Workspaces have no login of their own. You log in once as the owner, then enter a workspace by typing that workspace’s master password.

You are asked for it every time you switch. That is deliberate: the password unlocks the workspace’s stored credentials, so holding it open indefinitely would weaken the separation it exists to create.

Two separate mechanisms, worth understanding because they have different strengths:

Data separation — always active, everywhere. Every stored item belongs to exactly one workspace and is only ever read back for that workspace.

Filesystem confinement — Linux only. Agent processes are additionally confined by the kernel to their own workspace’s directory. Even a badly behaved agent cannot reach outside it.

Agents run on a schedule without you being present, which means Rookery needs to unlock a workspace’s credentials while nobody is entering the password. It stores what it needs for this, encrypted, so scheduled runs work on a machine that reboots unattended.

The trade-off is honest and worth stating: an installation that can run agents at 3am is an installation that can decrypt those credentials at 3am. The protection is on the machine and the account, not on a password held only in your head.