Can a human take over the conversation?

Yes. A handoff can carry conversation context, evidence and unresolved action state so the next person does not need to reconstruct the case from scratch.

Does every action happen automatically?

No. The same action-level governance can allow, review or block an external action regardless of whether it was requested through voice, browser or another execution surface.

Is one voice runtime always used?

No. The appropriate path depends on provider capabilities and whether the workflow needs explicit tool use, policy evaluation and business-state handling.

Voice agents · conversation to controlled action

Connect the conversation to the work that needs to happen next.

Q: Can a voice agent use business tools during a call?

Yes, when the configured runtime and tool permissions allow it. Tool-bearing flows should preserve explicit policy and outcome handling rather than hiding side effects inside the audio layer.

Createting treats voice as an execution channel on the same agent runtime: understand the caller, load approved context, propose or execute permitted actions, pause at policy boundaries and carry relevant context into a human handoff when needed.

Assess a voice workflow Explore the agent platform

Configured conversation simulation. It demonstrates runtime and governance mechanics; it is not a live call, latency benchmark or resolution-rate claim.

Call session · configured example

PIPELINE · TOOL-BEARING

Duplicate charge Reschedule appointment Sensitive cancellation

CallerI was charged twice for this month. Can you refund the duplicate €84?

Voice agentI can check the account and the two charges before I change anything.

Voice agentI found two matching charges. I can prepare the €84 refund, but this configured action needs review before it is submitted.

Voice agentThe reviewed refund was submitted. I can also send the confirmation to your email on file.

TranscriptWaiting for caller turnidle

ContextNo account context loadedpending

Action / policyNo tool action proposedpending

OutcomeNo downstream effectpending

Conversation continued with the verified result.

The channel stays connected to the same business context instead of ending at “the tool call returned.”

01Listen 02Load context 03Act / hand off 04Continue with result

Direct answer

What is a voice AI agent?

A voice AI agent is an agent that can understand and respond during a live audio conversation while using permitted business context and runtime capabilities to complete or coordinate work beyond the call itself.

The useful distinction is not “a chatbot that speaks.” A business voice workflow also needs session state, interruption handling, approved knowledge, tools or actions, policy boundaries, downstream verification and a clear human handoff when the agent should stop owning the decision.

Same runtime, different channel

Voice should not become a second automation stack beside your agents.

The current architecture treats voice sessions as a capability connected to the broader agent runtime rather than a disconnected phone-bot product.

Conversation in. Governed execution out.

The channel carries a live dialogue, but the business work still depends on the same questions: what context is available, which action is permitted, whether review is required and what result should be returned to the caller.

Audio / sessionChannel, session state and interruption-aware turn handling.

ContextApproved knowledge, workspace data and business-system state relevant to the conversation.

Agent runtimeReasoning, model routing and the same bounded capability model used outside voice.

Policy / toolsPermit, review or deny consequential actions instead of letting the call bypass control.

Continue / handoffUse the verified result in the next spoken turn or transfer ownership with relevant context attached.

Runtime modes

Choose the voice path around tool needs and runtime constraints—not around a marketing label.

The current Voice runtime exposes mode and channel decisions explicitly. Availability and provider bindings are runtime configuration, not something the website should pretend is universal.

PipelineCanonical path for tool-bearing business calls.

Keeps speech processing and the agent/tool runtime separable so business actions, policy and session behavior stay observable.

S2S / bounded modeSurface capability and availability honestly.

The current dashboard models S2S as a bounded mode with explicit availability and constraints rather than silently falling back for tool-bearing calls.

Human handoff

A good handoff transfers ownership without erasing the conversation.

Sensitive, uncertain or commercially important cases should be able to leave agent ownership without forcing the customer and operator to reconstruct the entire call.

Before handoff

Collect only the relevant caller intent, verified account context, actions already attempted and the reason the agent is escalating.

After handoff

Give the operator enough context to continue the case while keeping the decision itself with the human when policy or judgement requires it.

Where voice fits

Use voice when the conversation itself is part of the operating workflow.

The best use cases combine conversational demand with structured context, bounded actions and a clear outcome or handoff.

Inbound serviceAnswer, resolve or route customer requests.

Use approved knowledge and permitted account actions, then escalate cases that need judgement.

SchedulingTurn spoken intent into a verified booking state.

Check availability, apply the configured booking policy and continue the call from the actual result.

Qualification & routingCollect context before the right human or workflow takes over.

Use the conversation to enrich the next step instead of treating every call as an isolated voice transcript.

Current boundary

Voice quality is deployment-specific, and this page does not turn latency budgets into performance claims.

The runtime tracks voice session state and latency-related metrics, but actual end-to-end behavior depends on channel, providers, models, network conditions, tools and the configured deployment. This page therefore does not claim a universal response time, interruption latency, call capacity or resolution rate.

FAQ

Questions before connecting voice to business actions.

Can a voice agent use business tools during a call?

Yes, when the configured voice mode, agent capabilities and policy allow it. Tool-bearing calls should keep the business action and its result observable rather than treating speech generation as the entire workflow.

Can a voice action require human approval?

Yes. Voice does not bypass the shared action boundary. A consequential action can pause for review while the call session and context remain available for the next step.

What happens when the agent should not make the decision?

The workflow can hand ownership to a person and carry relevant conversation and business context forward so the operator can continue without reconstructing the case from scratch.

Does Createting guarantee a fixed voice latency or resolution rate?

No. End-to-end performance depends on the configured channel, providers, models, network, tool path and deployment. Public performance claims require a defined benchmark methodology and measured environment.

Start from the call outcome

Map what the caller wants, what context is required, which actions are allowed and where a human should take over.

Bring one real call flow and separate conversation, knowledge, tool actions, approval points, handoff and downstream evidence before choosing the final voice architecture.

Assess the workflow Architecture review

Connect the conversation to the work that needs to happen next.

What is a voice AI agent?

Voice should not become a second automation stack beside your agents.

Choose the voice path around tool needs and runtime constraints—not around a marketing label.

A good handoff transfers ownership without erasing the conversation.

Use voice when the conversation itself is part of the operating workflow.

Voice quality is deployment-specific, and this page does not turn latency budgets into performance claims.

Questions before connecting voice to business actions.

Map what the caller wants, what context is required, which actions are allowed and where a human should take over.

Let a voice agent act on the conversation—not just talk through it.

A voice agent is a channel into the same governed business runtime.

Automate the bounded action. Escalate the decision that should stay human.

Refund with approval

Appointment scheduling

Sensitive cancellation

Choose the voice path around the job the agent must perform.

Escalation should carry the conversation state with it.

Voice quality and telephony behavior are deployment-specific.

Questions before a voice agent acts on a live conversation.

Map what the agent may say, what it may change and when a human must take over.