QuoxVoice · voice for governed agents
Talk to your governed AI agents, out loud.
Real-time voice conversations with your Quox agents, right in the dashboard. Fullscreen, bound to the agent you choose, and witnessed like every other governed action.
Live today
What is live.
These work now, in the QuoxCORE dashboard. Everything below this section is labelled Soon or Planned, and means exactly that.
Voice in the dashboard
Real-time voice conversations with your agents, in the browser. No phone number, no separate app: open the voice surface and talk.
Fullscreen, per agent
The voice surface binds to one agent. You talk to Vulcan, or CommanderQ, or a specialist you built, and it answers as itself, not as a generic assistant.
Two agents behind one voice
A fast conversational voice keeps the line warm while it hands the real work to the agent’s fully tooled, governed self. Natural backchannels mean the line never feels dead while work runs.
Witnessed by default
Every session runs through the same governed execution layer as the rest of QuoxCORE and lands in the evidence trail. What was asked, what ran and what came back are all on the record.
Why two agents
A voice that talks while it works.
One model either answers fast and shallow, or thinks deeply and leaves you listening to silence. QuoxVoice splits the difference across two agents on one line.
The conversational agent is quick and warm: it greets you, works out what you want, and keeps things moving. When you ask for something real, it delegates to the doing agent, the same fully tooled, governed identity you would talk to in text, and narrates while that agent works, so the pause feels like a conversation rather than a loading spinner.
The doing agent is the governed one. It carries the tools, the permissions and the evidence, so a spoken request is held to the same standard as a typed one. Read more about that layer at the evidence layer and the platform it runs on at QuoxCORE.
Coming and planned
Voice everywhere you work.
What is not live yet, stated plainly. Soon means in build; Planned means on the roadmap and not yet started.
SoonAgents that phone you
When a long-running agent hits a real fork or finishes a job, it calls your phone, says which session it is calling on behalf of, and takes your decision. Proven in our own fleet today and hardening into a customer feature now.
SoonVoice channels in Discord
An agent holds a voice channel in your Discord server and converses there, so a call with your agents is one click from where your team already gathers. In build, not yet shipped.
PlannedYou call the fleet
Ring a number, verify yourself, and route straight to a live agent or session. Receptionist agents that answer your line, triage, and take a message.
PlannedMore places to talk
Wherever a channel can carry audio, the same governed voice can meet you there. Sequenced behind the live surface, phone, and Discord.
The difference
Every word, witnessed.
Every other voice agent sells you minutes. QuoxVoice gives you minutes you can prove. A voice that can act in your systems is only safe if you can see what it did.
Every call becomes one tamper-evident record: who was told they were speaking to AI, what they consented to, what was said, what the agent ran, and any opt-out, unified and queryable, not stitched together across separate vendor systems after the fact.
Built for the rules AI calling now lives under: disclosure at the start of the call, honoured opt-outs, do-not-contact scrubbing, and a per-call audit trail, by design. What an auditor can check for themselves lives at for auditors, the open protocol underneath at the WARD docs, and our own witnessed record at /security.
Ready when you are
Have a word with your agents.
QuoxVoice ships inside QuoxCORE. Open the dashboard, pick an agent, and start talking.
Live today: voice in the dashboard. Phone calls and Discord voice are labelled Soon and Planned above, and are not shipped to customers yet.