Route commands to agents by name
Type or say tell Alex to run the tests in the command bar and Alex’s window receives exactly run the tests — the addressing is stripped, a meteor animation flies to the receiving window, and the agent knows it’s Alex (every task carries a one-line identity context). Routing works for every window type, not just agents.
The addressing forms
Section titled “The addressing forms”| You say | Who gets it |
|---|---|
tell Alex to <task> / ask Alex <question> |
Alex |
Alex, <task> / Alex: <task> |
Alex |
<task> (no name) |
The focused agent window |
send <message> to Alex |
Alex |
send command to the terminal <command> |
The terminal (aliases: browser, kanban, music, whiteboard, code, hermes/ninja) |
hey hermes, <request> |
The Hermes orchestrator |
Dictation fillers are fine: “Uh ask Alex what the build status is” routes and strips correctly. Negations are respected — “Don’t ask Alex about this” is never treated as an address.
Good to know
Section titled “Good to know”- A minimized window still receives routed commands — it’s restored first, so nothing is lost.
- If nothing matches (no window by that name, no command), you get a toast instead of silence.
- Every route is recorded in the Audit Log.
- Voice and typing are equivalent everywhere: dictation fills the same command bar.
If the agent seems confused about its own name
Section titled “If the agent seems confused about its own name”Update to v0.1.80+ — earlier versions could pass the raw addressed sentence through, making the agent respond as if “Alex” were a stranger. Current versions strip the address and tell the agent who it is.