AI & Automation

AI Agents

Create a configurable AI assistant that handles customer messages across your connected channels — WhatsApp, Instagram DMs, Facebook Messenger, and post comments.

What is an AI Agent?

An AI Agent is a configurable assistant you train on your brand knowledge. When customers send messages to your connected social accounts, the agent reads the message, searches your approved brand knowledge (products, services, FAQs, policies, locations), and generates a reply.

Depending on the response mode you choose, the agent can:

Suggest a reply for your team to review before sending.
Queue a reply for an Admin or Owner to approve before it is sent.
Send automatically when confidence is high and all safety rules pass.
Stay disabled for a channel while still classifying and tagging conversations.

Plan requirement: AI Agents are available on the Business plan and above. Basic agent tools (brand knowledge search, product search, location, opening hours) are included. Advanced tools (lead capture, conversation tagging, human escalation) require the Professional plan or above.

Supported channels

You can assign an AI Agent to any combination of the following channels:

Channel	Message type
WhatsApp Business	Direct messages (DMs)
Instagram	Direct messages (DMs)
Facebook Messenger	Direct messages (DMs)
Instagram	Post comments
Facebook	Post comments

Creating an AI Agent

Who can do this: Owner, Admin. Navigate to: Manage → AI Agents.

Go to Manage → AI Agents in the sidebar.
Click Create agent.
Optionally pick a template (General Assistant, Sales Assistant, Support Assistant, or Lead Qualification Assistant) to pre-fill the agent settings.
Fill in the Agent name and Objective — the objective is the most important field; it tells the agent what it is trying to accomplish.
Set the tone (professional, friendly, casual, formal, energetic) and language(for example, en for English).
Choose a response mode (see below).
Select the channels this agent should handle.
Select which knowledge sources the agent can search (Products, Services, FAQs, Policies, Locations, Offers, or All).
Enable the allowed tools you want the agent to use.
Configure escalation rules so the agent knows when to hand off to a human.
Click Create agent.

Business Profile must be populated before creating an agent. The agent searches your brand knowledge to generate accurate replies. If your Business Profile is empty, the agent will have nothing to search and will give generic responses. Fill in your products, services, FAQs, and policies first. Go to Manage → Business Profile.

Response modes

Suggestion

The agent generates a reply and adds it as a suggested reply in the Inbox. A team member reviews and sends (or edits) it manually. No message is sent automatically.

Approval required

The agent generates a reply and it is queued for an Owner or Admin to approve before it is sent. Useful for regulated or sensitive accounts.

Automatic

The agent sends the reply automatically when: confidence is at or above 0.7, risk level is LOW, the conversation is not escalated, and your workspace compliance policy allows auto-sends. In all other cases it falls back to Suggestion mode.

Disabled

The agent does not generate or send replies on this channel. It will still classify conversation intent and apply tags if those tools are enabled.

Agent tools

Tools are actions the agent can take while handling a conversation. Enable only the tools that match the agent's role.

searchBrandKnowledge

Basic

Searches all approved brand knowledge in your Business Profile to answer customer questions.

searchProducts

Basic

Searches your AI-approved product catalogue to answer product queries.

getBusinessLocation

Basic

Retrieves approved business location details (address, city, country, phone) to answer location queries.

getOpeningHours

Basic

Retrieves approved opening hours to answer availability questions.

tagConversation

Advanced

Applies a label or tag to a conversation automatically based on context (e.g., "sales", "support", "complaint").

createLead

Advanced

Creates a sales opportunity in the Leads pipeline when a conversation qualifies as a genuine commercial enquiry.

updateLead

Advanced

Updates an existing lead linked to the current conversation with new information collected during the chat.

saveCustomerInformation

Advanced

Saves validated customer data (name, phone, email) to the conversation memory for your team.

requestHumanAssistance

Advanced

Triggers an escalation to a human agent and marks the conversation as requiring attention.

Basic tools are available on any plan that includes AI Agents. Advanced tools require the Professional plan or above.

Escalation rules

Escalation rules control when the agent stops handling a conversation and flags it for human attention. A conversation that is escalated appears in the Inbox with an escalation indicator and is moved to Human Active state — the agent will not auto-reply until a team member releases the conversation.

Low confidence

Escalates when the agent's confidence score falls below your configured threshold (default 0.4). Lower the threshold to escalate more often; raise it to let the agent handle more on its own.

Customer requests a human

Escalates when the agent detects phrases like "speak to a person", "real agent", "manager", or "representative".

Complaints

Escalates when the conversation is classified as a complaint.

Payment or refund disputes

Escalates when the conversation is classified as a refund request or payment dispute.

Working hours

You can set a working hours window (start and end time) for each agent. Outside of working hours the agent will not auto-send replies — it will still generate suggestions for your team to review when they come online. Leave working hours blank for 24/7 operation. Times use your browser's timezone when saved.

Workspace messaging safeguards

At the top of the AI Agents page you will see the Workspace messaging safeguards panel. These are workspace-level controls that apply to all agents and override individual agent settings:

Automatic sends — master switch for all auto-send behaviour. Disable this to stop all agents from sending automatically, regardless of their individual response mode setting.
Comment replies — controls whether agents can auto-reply to post comments.
Private replies — controls whether agents can send private replies in response to comments.

Platform-wide Meta policies (such as messaging windows and spam limits) always apply and cannot be overridden by workspace settings. Ensure your agents respect the 24-hour messaging window for WhatsApp and Messenger, and the 24-hour Instagram DM window.

Agent analytics

Each agent card on the AI Agents page shows live summary metrics:

Handled — total conversations the agent has processed.
Escalations — conversations escalated to a human.
Leads — sales opportunities created by the agent via the createLead tool.

Click Activity on any agent card to see a detailed activity log for that agent, including the type of action taken and when.

Testing an agent safely

Click Test on any agent card to open the test sandbox. Type a customer message and click Run test. The agent generates a response using your real brand knowledge, but the sandbox never sends any messages or creates real leads. Use this to verify the agent responds correctly before enabling automatic sends.

Built-in templates

When creating a new agent, you can start from one of four pre-configured templates. Templates pre-fill the agent name, role, objective, and recommended tools — you can customise all fields after selecting a template.

General Assistant

Answers customer enquiries accurately using approved brand knowledge. Uses the four basic knowledge tools.

Sales Assistant

Turns relevant product and service enquiries into qualified opportunities. Adds createLead, tagConversation, and saveCustomerInformation tools.

Support Assistant

Resolves support questions from approved information and escalates cases requiring human attention. Adds tagConversation and requestHumanAssistance tools.

Lead Qualification Assistant

Collects only the missing information needed to qualify genuine commercial enquiries for human follow-up. Adds createLead, updateLead, and saveCustomerInformation tools.

AI credits and agents

Each conversation processed by an AI Agent consumes AI credits from your monthly allowance. The number of credits used depends on the length and complexity of the conversation. Credits are deducted when the agent generates a reply — not when the reply is sent. When your credit balance reaches zero, agents stop generating replies and fall back to an empty suggestion for your team to fill in manually.

You can purchase additional AI credit top-up packs from Settings → Billing. Top-up credits do not expire. See the AI Credits guide for details.

Related guides

💬

Inbox

How to manage conversations, assign to team members, and approve AI replies.

✨

AI Features

Caption generation, hashtags, rephrase, and translation in the Composer.

🪙

AI Credits

Understand your credit balance, top-ups, and history.

🔗

Social Accounts

Connect WhatsApp, Instagram, and Facebook before enabling an AI Agent.