Hands-free, still governed

Talk to your agents — approvals still require a yes.

A natural-language voice surface over OpenAI Realtime, with ephemeral client secrets so the API key never reaches the browser, and the same approval rules still in force.

Voice AI is convenient but usually ungoverned and leaks credentials to the browser. The platform's voice orchestrator runs over OpenAI Realtime, and the server mints ephemeral client secrets so the OpenAI API key never reaches the browser.

Each agent has a voice identity, and spoken work is delegated to the same governed runs as everything else. Sensitive actions still require an explicit spoken or tapped yes — there is no silent auto-approve — so talking to an agent is just as accountable as clicking through a plan card.

Voice models can be a cost black box until the bill arrives, so the surface offers Economy, Balanced, and Premium cost/quality profiles with per-session budgets and a live cost meter. You choose a tier and watch spend in real time during a session, keeping voice cost predictable and aligned to the value of each interaction.

In this capability

What's included.

Natural-language voice surface over OpenAI Realtime.
Server mints ephemeral client secrets — the OpenAI API key never reaches the browser.
Work is delegated to governed runs; sensitive actions still require an explicit yes.
Economy, Balanced, and Premium cost/quality profiles with per-session budgets.
A live cost meter so voice spend is visible in real time.

Permission, policy and audit on every tool call · ← Back to all features · Read the docs →

See governed agents in your own product.

Book a demo and we'll walk your team through the governance model, the two-seam integration, and a use case for your industry.