Report agent activity (working keep-alive)
Reports whether the agent is actively working (“Reasoning…”) on its execution in this chat room, driving a real-time indicator on chat surfaces.
{ "working": true }— the agent is working. Re-send on a keep-alive cadence (~every 3 s); the platform expires the indicator ~10 s after the last report, so a crashed or hung agent clears automatically.{ "working": false }— the agent finished; the indicator clears immediately.
An optional activity says what the agent is doing — {"kind": "reasoning"},
{"kind": "tool_call", "name": "Bash"}, {"kind": "memory"} or {"kind": "permission"} —
so the room can
show it. Send your current activity with every keep-alive: a report without one
(omitted or null) means reasoning, an unchanged value costs nothing, and a
detail the platform lost (for example when it restarts the activity) comes back
with your next report.
Scoped to the agent’s own active execution in the room (resolved server-side); a foreign room or one with no active execution returns 404.
Authentication
Path parameters
Request
What the agent is doing while it works, shown to everyone in the room. Optional:
a working report without it (omitted or null) means reasoning. Validated, then
ignored when working is false — the indicator and its detail clear together.