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

# AI Agents

> Create AI agents that handle conversations, follow your instructions, and hand off to your team.

## What is this?

AI Agents (beta) lets you create AI-powered agents that automatically handle conversations on your connected channels — answering questions from knowledge sources, taking actions like updating contacts or handing off to a human, and following the instructions and guardrails you define, without needing a manual automation flow for every scenario.

\[SCREENSHOT: ai-agents-list — The AI Agents list page showing the Total/Active/Inactive stat tiles, Agentic Mode toggle, and the agents table]

## What can I do here?

* See all AI agents for the active channel with **Total Agents**, **Active Agents**, and **Inactive** stat tiles you can click to filter.
* Search agents by name or description, and filter by type (**Support**, **Sales**, **FAQ**, **Booking**, **Custom**).
* Toggle **Agentic Mode** on or off for the channel from the top toolbar. This is the channel's master switch: when it's off, no agent answers incoming messages, even if the agent itself is **Active**.
* Create a new agent (subject to your plan's agent limit, shown as a counter on the Create button).
* For each agent in the table: **Edit**, **Test** (opens the agent builder), **Export** (downloads the agent as a JSON file), **Activate/Pause**, and **Delete**.
* Select multiple agents to bulk-select and delete.
* **Import** an agent from a previously exported JSON file into any channel you choose.
* Refresh the agent list.

## How to use it

<Tip>
  New to AI agents? Follow [Create Your First AI Agent](/workspace/ai-agents/create-your-first-agent) — a step-by-step guide with a full example.
</Tip>

1. Go to **AI Agents** in the sidebar (`/ai-agents`). You need an active channel connected to WhatsApp, Instagram, Facebook, or Web Chat.
2. Click **Create** to start a new agent (or **Edit**/**Test** on an existing row to open the same builder). The builder has one tab for each area of the agent.

\[SCREENSHOT: ai-agent-detail-tabs — The AI agent builder page showing the full row of tabs across the top]

<Tabs>
  <Tab title="General">
    Set the Agent Name, Agent Type (Custom, Support, Sales, FAQ, Booking), and a Description (with an "AI Rewrite" button to have AI polish your description). Write the description clearly — when a channel has more than one active agent, the platform uses each agent's description to decide which one should answer a message.

    Save the agent first: the Capabilities, External Capabilities, Skills, and Workflows tabs can only be configured after the agent has been created.

    \[SCREENSHOT: ai-agent-general-tab — The General tab with Agent Name, Agent Type, and Description fields]
  </Tab>

  <Tab title="Behavior">
    Set the **Tone** (Neutral, Friendly, Professional, Casual, Formal), **Priority** (1–1000 — higher-priority agents are triggered first when several could answer), and **Status** (**Active**, **Inactive**, or **Testing**). Use **Testing** while you're still tuning an agent.
  </Tab>

  <Tab title="Instructions">
    Write the System Instructions that tell the agent how to behave and respond, with an "AI Rewrite" helper button.

    \[SCREENSHOT: ai-agent-instructions-tab — The Instructions tab with the system instructions textarea and AI Rewrite button]
  </Tab>

  <Tab title="Knowledge">
    Click **Add Source** to give the agent reference material — Manual Text, a Website URL, or a file (PDF, Word, Markdown, Text, CSV, Excel), plus Sitemap Crawl and llms.txt Crawl. Each source shows a processing status (Pending, Processing, Processed, Failed).

    \[SCREENSHOT: ai-agent-knowledge-tab — The Knowledge tab showing the New Knowledge Source form and the list of added sources with status badges]
  </Tab>

  <Tab title="Capabilities">
    Toggle built-in actions the agent can take, grouped by Contacts, Conversation, Knowledge, Catalog, Events, and Automation — for example updating a contact, handing off to a human, or searching the catalog. Enabling a capability can require customer confirmation before it runs, and the "hand off to human" capability lets you set an assignment mode (Unassigned, Fixed team member, Round robin, Least busy).

    \[SCREENSHOT: ai-agent-capabilities-tab — The Capabilities tab with a capability enabled and its confirmation/handoff-assignment options expanded]
  </Tab>

  <Tab title="External Capabilities">
    Click **Add External Capability** to connect custom HTTP requests or MCP tools the agent can call. This requires the `automations:edit` permission.
  </Tab>

  <Tab title="Skills">
    Click **Add Skill** to attach reusable skills to the agent.
  </Tab>

  <Tab title="Workflows">
    View automation sequences bound to this agent. This is a read-only list — the workflows themselves are configured in [Automation](/workspace/automation/overview).
  </Tab>

  <Tab title="Channels">
    Choose which communication channels this agent applies to — All Channels, WhatsApp, Facebook, Instagram, or Web Chat. Choose **All Channels** to let the agent run on every channel type.

    This tab also holds a few settings for how the agent talks:

    * **AI Setting Source** — choose which AI setting (provider and key) powers this agent. If none are listed, add one first under [Settings → AI Setting](/workspace/settings).
    * **Allowed Response Modes** — **Text** is always on. Turn on **Buttons** and **List** to let the agent reply with interactive buttons and lists.
    * **Response Language** — **Auto** (reply in the customer's language) or **Fixed** (always reply in a language you choose, such as Spanish or French).
    * **Media Capabilities** — turn on **Image Understanding**, **Voice Input**, **Voice Output**, and **Video Understanding** for the media types this agent can handle.

    \[SCREENSHOT: ai-agent-channels-tab — The Channels tab showing the channel selector pills]
  </Tab>

  <Tab title="Guardrails">
    Toggle **Enable Guardrails** to add safety checks that prevent harmful, biased, or inappropriate responses. The tab explains what guardrails do and why to keep them on.
  </Tab>

  <Tab title="Test">
    Use the Agent Playground to send real test messages to the agent and see its replies — and which tools it used — before you turn it live. Start a **New Conversation** to reset the test session.

    \[SCREENSHOT: ai-agent-test-playground — The Test tab's Agent Playground with a sample conversation]
  </Tab>
</Tabs>

### Export and import an agent

Back on the list page, use **Export** to download an agent's configuration as JSON, and **Import** to bring that JSON into another channel — this creates brand-new agent, source, and knowledge records rather than overwriting anything.

## Related docs

* [Create Your First AI Agent](/workspace/ai-agents/create-your-first-agent)
* [Automation Overview](/workspace/automation/overview)
* [Inbox](/workspace/inbox/overview)
* [Analytics](/workspace/analytics) — AI Usage tab