Feature request for product/service
Cursor IDE
Describe the request
I use live voice calls with the primary (parent) Agent (two-way TTS, e.g. Grok voices like Ara) while also running child / sub-agents spun up from that session (Task subagents, tiled Agents Window, multitask workers, background agents). Today, parallel work is easy to see in the UI but hard to hearâupdates often sound like one undifferentiated voice.
Parent vs child voice routing (summary): The primary agent keeps one default voice (e.g. Ara) for the user’s main live call. Every child agent created from that primary session should get its own voiceâby default randomly picked from an allowed set, never the parent’s voice, and distinct from sibling children.
Reality check: This may be partly unrealistic with how voice mode works today. The voice layer often summarizes multiple agents’ status in a single sentence rather than compartmentalizing output per agent, so distinct voices per child may not map cleanly onto how responses are actually delivered. Still, it’s a fun concept worth exploringâespecially if product direction moves toward clearer per-agent audio identity.
Forum research (Oct 2026) did not find an existing thread that covers primary + per-child-agent voice routing exactly. Closest topics are per-Grok-Bot voice settings (and related persistence bugs), Voice-First Agent Supervision, and Full Conversational Voice Mode.
1) Primary voice for the main (parent) agent
Allow the user to set a primary / default voice for the main parent Agent voice session (e.g. Ara), at account or workspace level, so every new live call with the primary agent starts with that voice unless overridden.
This is the voice for the parent agent the user is talking to on the live callânot the voices of sub-agents spawned underneath it. It is separate from picking a voice per Grok Bot profile; it should apply to the primary IDE Agent voice-call experience the user treats as “the lead developer.”
2) Distinct random voice per child / sub-agent (spawned from primary)
When the primary agent spins up child agents (subagents, multitask, separate Agent tiles, background workers tied to that parent session), assign each child a different TTS voice automatically. Default behavior: randomly select a voice from an allowed pool when the child is created, ensuring:
- the child never uses the primary parent’s default voice (e.g. not Ara if that is the parent);
- each sibling child gets a distinct voice when possible.
Optional: let users override with explicit voice mapping per agent or subagent label.
Goal: the user hears who is speakingâparent vs each childâwithout looking at the screen (when the audio pipeline supports it).
Why this matters
- Parallel agents are a core workflow; voice is increasingly used for hands-free supervision.
- Per-agent voice coud reduce cognitive load when multitasking (walk, commute, multiple tiles)âif delivery can be separated per agent.
(Related forum threads: see forum search for Full Conversational Voice Mode, Voice-First Agent Supervision, Grok Bot voice settings.)
Operating System (if it applies)
MacOS