Voice is one more channel of the same agent.

The agent on the call is the same one on WhatsApp and chat: same rules, same audit trail, same cost ceiling.

A call is an actor.

Every call is a supervised actor with a turn state machine: listening, thinking, speaking. Barge-in cancels the model mid-sentence, and hold and transfer are states.

Listening

Audio arrives over telephony or the browser and becomes a transcript, mandatory on every call.

Thinking

The agent checks the rule and the authorized systems, under the same guardrails as every other channel.

Speaking

The voice speaks while the model writes: synthesis streams out without waiting for the full answer.

Barge-in

When the customer interrupts, barge-in cancels the speech and the cancellable work of the turn. What was already confirmed keeps its state.

Hold and transfer

Holding and transferring are states of the call, not exceptions, and they stay in the trail like any other step.

Voice is not another project.

In a framework, voice is another project. In a managed platform, another product. In Tessera, the call inherits the rules, guardrails, trail and cost of the same agent.

WhatsApp, voice, API and queue come in through the same edge, with identity, tenant, and PII and policy filters before the first step.

Two engines, one driver.

Cascade (speech, text, agent, text, speech) and native speech-to-speech under a single dialogue driver, with a mandatory transcript. The engine can be swapped mid-call.

A human listens, whispers or takes over.

Listen

The supervisor follows the call live from the supervisor console.

Whisper

The supervisor coaches the agent during the call without taking the conversation away from it.

Take over

A warm or cold handoff carries the redacted transcript and the collected data; the disposition goes back to the CRM.

Consent is an aspect, not a checkbox.

Recording consent, brand voice and tone are policies woven into the agent by scope: versioned, enforced and audited like every other rule.

The rule outside the model

The decision on the call comes from a rules engine, versioned and explained, not from the prompt.

Every call in the trail

Who asked, which rule, who approved and why, hash-chained and reproducible.

The same cost ceiling

Every step of the call feeds the trace, the cost ledger and the trail of the same agent.

Connected to telephony and to your systems.

Dialogue drivers, call runtime, transcription and synthesis ports, WebRTC, telephony media streaming and the supervisor console are already delivered. Telephony and WhatsApp are configured per deployment, in the institution’s own cloud.

The journey changes. The kernel doesn't.

Journeys change from one industry to the next, from banking to telecom, insurance, utilities and healthcare. The kernel and the governance of the call do not.

Voice and governance, under one kernel.

Govern what you've already built.

Connect agents from different frameworks to a common layer for operations and governance.

By submitting, you agree to receive the Tessera newsletter, as described in the Privacy Policy. You can unsubscribe at any time.