# Chat Agents

Chat Agents are AI-powered website assistants that engage visitors in real time, answer questions, collect lead information, and support handoff to human operators when appropriate.

## Key Concepts

- **Chat Agent** — The AI configuration that powers a specific chat widget
- **Widget** — The embedded chat interface that appears on your website
- **Chat Session** — A single conversation between a visitor and the AI
- **Operator Panel** — A dashboard for human agents to monitor and take over chats

## Getting Started

1. Navigate to **Agents > Chat Agents**
2. Click **+ Create Chat Agent**
3. Configure the agent name, instructions, and appearance
4. Save and embed the widget code on your website

## Chat Agent List

The Chat Agents page gives you a central place to manage every chat experience in your workspace. For each agent, you can review:

- **Name** — Agent identifier
- **Status** — Active or paused
- **Session Count** — Total chat sessions
- Quick actions: view, edit, view chats

<!-- SCREENSHOT: chat-agents-list.png — Chat agents list page -->

## Live Preview

The chat agent editor includes a live preview panel that runs against the **real** chat engine, not a stripped-down stub. Each preview turn uses a **sandbox copy** of the agent that includes your unsaved edits, so replies have full production fidelity — the agent's tools, RAG / vector stores, MCP servers, and memory all behave exactly as they will once published.

A live **tool-call inspector** shows each tool the agent calls as the reply streams in, with the parameters sent and the result returned expanded inline, so you can see *why* the agent answered the way it did. Reset the conversation at any point to start fresh; a brand-new agent gets its sandbox draft created on the first message, so your preview conversation survives when you save.

## Agent Handoff

A chat session can be handed off from one chat agent to another so a conversation moves to the right specialist without losing context. In the **Agent Handoff** section of the chat agent form:

1. Enable handoff for the agent.
2. Pick the target agents it may hand off to.
3. Describe the **circumstances** for each target so the AI knows when to transfer.

When a handoff happens, ownership of the session is reassigned to the target agent, a system message is written into the transcript, and the transfer is recorded. Because each target carries its own handoff policy, a single conversation can move through a chain of agents, and the full handoff history is available on the session transcript. An operator can also trigger a handoff manually.