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

# Customer-facing agents

> How customer-facing agents answer contacts from an inbox, how Autopilot and Copilot modes work, and where to test and review them.

Customer-facing agents (called **customer agents** in Conduit) answer your contacts: guests, leads, customers and anyone else who messages you. Each one listens to one or more inboxes and replies to the messages that arrive there, using its instructions, Knowledge, Skills and tools.

Use customer-facing agents for support, sales, guest messaging, lead intake and any other job where the AI talks directly to a contact. For work that happens behind the scenes, use an [internal agent](https://docs.conduit.ai/build-your-agent/agents/internal-agents/overview.md) instead.

## How they differ from internal agents

| | Customer-facing agent | Internal agent |
| - | - | - |
| **Talks to** | Contacts, in their conversation | Your team and your systems |
| **Starts from** | Messages arriving in its inbox | Messages, schedules, webhooks, platform events, chats |
| **Escalation rules and behavior** | Yes | No |
| **Sandbox testing** | Yes | No. Every run is real. |
| **Credits** | Doesn't consume credits | Consumes credits |

## Create a customer-facing agent

**1. Open Agents**

Go to **Agents** in the workspace sidebar.

**2. Add a customer agent**

Select **Add agent > Create customer agent**. Conduit also creates an inbox for the agent, so inbound conversations for that contact type can be routed to it.

**3. Configure it**

Fill in the **Persona**, **Tools**, **Skills** and **Settings** tabs. See [Agents](https://docs.conduit.ai/build-your-agent/agents) for what each tab holds.

## Inbox triggers

A customer-facing agent runs when a message arrives in an inbox it's attached to. The agent's sidebar lists them under **Inbox triggers**, and warns you when another agent already listens to the same inbox.

Messages in the **Unmapped** inbox don't trigger agents. If an agent is attached to **Unmapped**, Conduit flags it so you can remove the trigger or route those messages to another inbox. See [Triggers](https://docs.conduit.ai/build-your-agent/agents/triggers.md).

## Autopilot, Copilot and Inactive

Set how the agent responds from the **Agent behavior** menu in its top bar. Pick a mode, then select **Save**.

![The Agent behavior menu with Autopilot selected, followed by Copilot and Inactive, above a Save button](https://mintcdn.com/conduit-58d85182/Dxa79236R9TzscJw/images/customer-facing-agents/agent-behavior-menu.png?fit=max\&auto=format\&n=Dxa79236R9TzscJw\&q=85\&s=2b394e03b1937c362bebd7917edca8a7)

| Mode | What happens |
| - | - |
| **Autopilot** | The agent answers what it can and sends replies on its own. It escalates what it can't handle. |
| **Copilot** | The agent drafts every reply as a [proposed reply](https://docs.conduit.ai/inbox/agent/proposed-replies.md), and your team approves each one before it's sent. |
| **Inactive** | The agent is turned off. |

Most teams start in **Copilot**, review proposed replies until they trust the agent, then switch to **Autopilot**. See [Road to autopilot](https://docs.conduit.ai/build-your-agent/road-to-autopilot/overview.md).

## Settings that only customer-facing agents have

Because they reply to contacts, customer-facing agents have extra sections on the **Settings** tab:

* **Escalation rules**: what happens when the agent escalates a message.
* **Behavior**: language, voice, working hours, reply delay, auto mark-as-done and signature.

See [Agent settings](https://docs.conduit.ai/build-your-agent/agents/settings.md). For what the agent already knows about a guest's stay, see [Reservation context](https://docs.conduit.ai/build-your-agent/agents/customer-facing-agents/reservation-context.md).

## Test and review

* **Before going live**, use Sandbox to simulate conversations with realistic contacts and reservations. See [Testing agents](https://docs.conduit.ai/build-your-agent/testing-agents.md).
* **After publishing**, open the **Activity** tab. It lists one row per run with its status, contact, start time and disposition, and lets you open the trace to see the instructions, Knowledge and tools the agent used.

The metrics at the top of **Activity** show the agent's **Runs**, **Autopilot rate** and **Escalations**.

Filter runs by status, by disposition (such as **Resolved**, **Escalated** or **No op**), or to sandbox runs only. See [Understand dispositions](https://docs.conduit.ai/build-your-agent/skills.md#understand-dispositions) for what each disposition means.

## Voice agents

Voice agents also talk to contacts, but on phone calls. They're configured separately because they need phone number routing, voice behavior, transfer rules and call-specific guardrails. See [Agents](https://docs.conduit.ai/build-your-agent/agents#voice-agents).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.