Ir al contenido
Esta página aún no está traducida por completo a tu idioma. Algunas secciones se muestran en inglés.
Inbound

El asistente telefónico IA que solo dice lo que pone en tu web

El agente inbound de anicall.io atiende llamadas las 24 horas, responde a preguntas concretas exclusivamente con contenido de tu web, tus documentos y tu catálogo de respuestas, y reserva citas directamente en tu calendario.

Ver precios

Actualizado: noviembre de 2026

Límite de conocimiento «solo con fuentes»

Cómo funciona el límite de conocimiento

Con el nivel «solo con fuentes», el agente de anicall.io responde a preguntas concretas exclusivamente desde tu web, tus documentos y tu catálogo de respuestas – lo que no esté ahí se convierte en una laguna de conocimiento que tú apruebas.

Semáforo de cristal con luz verde, ámbar y roja: símbolo de la revisión de cada respuesta
  1. Importar tu conocimiento

    Tu web (rastreo de varias páginas), archivos PDF, DOCX y de texto, YouTube, RSS y tu catálogo de respuestas forman la base de conocimiento del agente.

  2. Responder solo con fuentes

    Antes de cada respuesta, anicall busca fuentes adecuadas. Si no hay, el agente lo dice con honestidad y ofrece devolver la llamada – en lugar de adivinar.

  3. Revisión tras la llamada

    Un segundo modelo comprueba cada respuesta frente a las fuentes: verde = respaldada, amarillo = pregunta abierta, rojo = afirmación sin fuente.

  4. Aprobar la laguna

    Las preguntas abiertas aparecen como lagunas en tu panel. Tu respuesta aprobada sirve como fuente desde la siguiente llamada.

Más niveles por agente: estricto, equilibrado, abierto.

Four levels of the knowledge boundary

Per agent, you decide how closely it sticks to its knowledge: grounded only, strict, balanced or open.

Grounded only

Factual questions only with content from website, documents and answer catalog. If knowledge is missing, the agent offers a callback and the question becomes a knowledge gap.

Strict

The agent sticks to its approved tasks and clearly declines other topics.

Balanced

A little small talk is fine, then the agent steers back to its core tasks.

Open

The agent also answers questions outside its core tasks.

Where the agent gets its knowledge

An anicall agent's knowledge base consists of your website, uploaded documents, YouTube and RSS sources, and your approved answer catalog.

Website

Multi-page import of your website – services, opening hours, prices, FAQ.

Documents

PDF, DOCX and text files such as price lists, service descriptions or house rules.

YouTube & RSS

Video transcripts and content from feeds such as your blog or newsroom.

Answer catalog

Your approved answers to knowledge gaps – available as a source from the next call on.

Appointments straight into your calendar

During the call, the agent checks free slots and books appointments in Google Calendar, Microsoft 365, CalDAV or Calendly.

Google Calendar

Connect with your Google sign-in; free slots and bookings directly in the selected calendar.

Microsoft 365

Outlook calendars via your Microsoft account.

CalDAV

For calendars that support the open CalDAV standard.

Calendly

The agent uses the times of your Calendly event type and books there. Requires a paid Calendly plan.

Transfer, callback and analysis

If a caller wants to speak to a human, the agent transfers live or takes a callback request.

Live transfer

Set a transfer number per agent; on request the agent transfers the call live.

Callback request

If nobody is available, the agent takes a callback request – including the callback.requested webhook.

Parallel calls

Several calls at once instead of a busy signal.

Transcript & summary

Every call is transcribed, summarized and stored searchably in your dashboard.

Preguntas frecuentes

Does the AI phone assistant make up answers?

With the “grounded only” level, the agent answers factual questions only with content from your knowledge base. If it finds no source, it says so and offers a callback. After the call, a second model checks every answer against the sources and shows the result as a traffic light.

What happens to questions the agent couldn't answer?

They show up as knowledge gaps in your dashboard; identical questions are counted together. You answer or dismiss them; approved answers go into the answer catalog and apply from the next call on.

Which calendars are supported?

Google Calendar, Microsoft 365, CalDAV and Calendly. Calendly requires a paid Calendly plan because booking runs through the Calendly Scheduling API.

Can I test the agent first?

Yes. You test the agent for free in the browser: create it, import your website and run test calls. For real calls you top up tokens – from €20.

Pruébalo con tu propia web

Prueba gratis en el navegador: importa tu web, haz una llamada de prueba y mira en la revisión qué respuestas estaban respaldadas.