Agent windows
Routing is great until you know exactly which agent you want. The dock at the bottom right gives every agent its own window - a direct line with no middleman.
narrated · sound on
Why this matters
Isolation is the quiet safety feature: what you tell one agent is not ambient background for another. Windows make context boundaries visible enough to trust.
The walkthrough
Open a direct line
Click any agent chip in the dock - Hermes, Claude, OpenClaw. A window opens with that agent's own conversation. What you type goes straight to it; Cronus does not route, rewrite, or supervise.
Windows behave like windows
Drag by the title bar. Stack as many as you like. Close one and its conversation is kept - reopen the chip and the transcript is right where you left it.
Live output, honest dots
While an agent works, its raw output streams into the window - useful for long jobs. A green dot on a chip means that agent is installed and reachable; gray means it is not on this machine, and the window will say so instead of pretending.
When to use which
Rule of thumb: when you do not care who does the job, ask Cronus. When you know exactly who you want and what to say, open the window. Both are first-class.