- Primeros pasos
- UiPath Conversational Agents
- Agentes de UiPath en Studio Web
- Acerca de los agentes de UiPath
- Licencia
- Ejecución de agentes
- Agentes y flujos de trabajo
- Mejores prácticas para construir agentes.
- Elegir el mejor modelo para tu agente
- Mejores prácticas para publicar e implementar agentes
- Mejores prácticas para el diseño de contexto.
- Prácticas recomendadas para DeepRAG y Transformación por Lote: JIT frente a estrategias basadas en índice.
- Trabajar con archivos
- Contextos
- Escaladas
- Evaluaciones
- Seguimientos de agente
- Puntuación del agente
- Gestión de agentes de UiPath
- Agentes de UiPath codificados
- Crear con agentes de codificación
Real-time voice conversational agents that handle incoming calls and place outgoing calls, built in Maestro Flow with the Voice agent node.
A voice conversational agent handles real-time phone conversations. Callers dial a number you own, and the agent can run tools during the call to read or update live data.
Cómo funciona
A voice agent runs on a real-time speech model. Audio goes straight to the model, with no separate transcribe-reason-synthesize pipeline in between. That is what makes the conversation fast enough for the caller to interrupt. Tools run outside that audio path, on the shared agent execution engine (Unified Runtime) that serves tool calls for every agent type.
You bring the phone number on a SIP trunk, or from your own Twilio account. The call reaches UiPath, which holds the model session for its duration and runs the agent's tools.
Voice conversational agents are built in Maestro Flow, not Agent Builder. There is no separate voice agent to publish or import.
- Build in Maestro Flow — modeling the call with the Voice agent, Incoming call, Create outgoing call, and End call nodes. Model, persona, prompt, and tools are configured on the Voice agent node.
- Deployment — connecting a phone number over SIP or your own Twilio account, and binding it to a published Flow.
- Observability — reading a call's transcript, tool calls, and timings in the run's execution trace.
Limitaciones
Estamos ampliando activamente el producto con nuevas capacidades. La siguiente tabla describe las limitaciones actuales, que se prevé abordar en las próximas versiones.
| Limitación | Descripción |
|---|---|
| Call duration | Calls are limited to 10 minutes |
| Call recordings | Audio recordings of calls are not available. |
| Hand off call to a person | Live call transfer to a person is not supported. |
| SDK support / chat surfaces | Voice agents are only available via phone calls |
| Automation Suite | Voice agents are available in Automation Cloud only. |
Preguntas frecuentes
Which phone providers work?
Voice agents work with the following phone providers:
- Twilio
- Amazon Chime SDK
- Genesys
- Five9
- any SIP providers that meet the SIP trunk requirements
Providers that do not work:
- Amazon Connect: use the Amazon Chime SDK instead.
- Azure Communication Services: unless you have a managed SBC you can use.
- Zoom
Which languages are supported?
Language coverage and quality depend on the selected model. For example, Gemini Live models support 24 languages. Accent comprehension and responding in different accents are limited by the model's capabilities.
Which guardrails and data protections apply?
UiPath enforces Zero Data Retention (ZDR) with all AI vendors, meaning no audio or transcripts are stored with our AI vendors such as OpenAI and Google.
Agent-level guardrails are not available for voice agents. This includes centralized guardrails in the AI Trust Layer, such as masking personally identifiable information (PII) and content policies. Tool-level guardrails are available on the tools connected to the Voice agent node. See Observability.