> ## Documentation Index
> Fetch the complete documentation index at: https://docs.recepta.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI agents

> The AI that answers your phone — what an agent is made of and how to configure one.

An agent is the AI that talks to your customers. It combines a voice, your business knowledge, per-caller-type instructions, and a set of actions it's permitted to take.

Your first agent is created automatically at signup. Most businesses never need a second one.

## What makes up an agent

| Component          | What it controls                              | Where to set it                                    |
| ------------------ | --------------------------------------------- | -------------------------------------------------- |
| Voice              | How it sounds, and its speaking style         | [Voices](/receptionist/voices)                     |
| Business profile   | Who it says you are, hours, service area      | [Business profile](/receptionist/business-profile) |
| Knowledge          | What it can answer                            | [Knowledge base](/receptionist/knowledge-base)     |
| Caller-type config | What it asks and collects, per type of caller | [Caller types](/receptionist/caller-types)         |
| Transfer rules     | When it hands off to a human                  | [Transfer rules](/receptionist/transfer-rules)     |
| Booking settings   | Whether and how it books                      | [Appointments](/booking/appointments)              |

## Configuration, not prompts

You configure an agent through structured settings rather than by writing prompt text. Fill in your hours, services, prices, and the questions you want asked, and Recepta.ai assembles the agent behavior from that.

The practical consequence: **improve the inputs, not the wording**. If the agent gives a wrong price, fix the price in your knowledge base. If it forgets to ask for an address, add that field to the relevant caller type. Reaching for clever phrasing is almost never the fix.

<Tip>
  When onboarding, pick the [industry template](/receptionist/industry-templates) closest to your business. It pre-fills sensible qualification questions for that vertical, which you then edit — much faster than starting empty.
</Tip>

## Running more than one agent

Add another agent when behavior genuinely needs to differ, not merely to organize things:

<Columns cols={2}>
  <Card title="Good reasons" icon="circle-check">
    A second [location](/account/locations) with different hours and service area. A different language. An outbound campaign agent with a distinct opening and goal.
  </Card>

  <Card title="Poor reasons" icon="circle-xmark">
    Separating services you already list in one knowledge base. Tracking — use call tags and analytics instead. Testing a wording change — edit the input instead.
  </Card>
</Columns>

Each agent is assigned to one or more [phone numbers](/phone/numbers). That assignment is what decides which agent answers a given call.

## The flow builder

For calls that need branching logic beyond the standard caller-type flows — multi-step qualification, conditional routing, structured data capture — the builder lets you lay out the conversation as an explicit flow with branches and actions.

Most workspaces don't need it. Reach for it when you've hit a genuine limit of the standard configuration, not before.

## Testing before you go live

Always call your own agent before pointing real traffic at it.

<Steps>
  <Step title="Run the happy path">
    Call as a new customer wanting your most common service. Book an appointment. Confirm it landed on the right calendar.
  </Step>

  <Step title="Run the awkward path">
    Ask about something you don't offer. Ask for a price on an unusual job. Interrupt mid-sentence. Give a partial address.
  </Step>

  <Step title="Run the escalation">
    Describe an emergency and confirm the transfer fires and reaches the right person.
  </Step>

  <Step title="Read the transcript">
    Open the call under [Call logs](/receptionist/call-logs). Reading is faster than listening, and mistakes are more obvious in text.
  </Step>
</Steps>

<Warning>
  Changes to an agent apply to calls that start after you save. A call already in progress finishes on the configuration it started with.
</Warning>