Creating and Configuring an AI Agent

You can create multiple AI agents per workspace, each with its own role, instructions, knowledge sources, and assignment behavior. To create one, go to AI Agent in the sidebar and click "Create Agent". The setup page is organized into tabs — Overview, Guidance, Escalation, Knowledge, and Automations — which this guide walks through in order.

Overview tab

Start here to set the agent's identity and high-level voice:

• Agent name — how the agent refers to itself in conversations
• Activate — the master switch that turns the agent on or off
• Role — a short description of what the agent specializes in (e.g. "You are a billing specialist handling payment, invoice, and subscription questions"). The role shapes how the agent interprets questions and what it treats as in-scope. Template chips (General support, Billing, Technical support, Onboarding, Sales) give you a starting point
• Tone of voice — Humorous, Friendly, Neutral, Matter-of-fact, or Professional
• Answer length — Concise, Standard, or Thorough

Guidance tab

The Guidance tab is where you shape how the agent answers. Each section holds a list of short instruction items you can add, edit, enable, or disable:

• Communication style — fine-tune the tone, style, and format of replies beyond the Overview defaults
• Context and clarification — when and how the agent should ask for more information
• Content and sources — which sources the agent may reference or share
• Scope and Boundaries — tell the agent how your business works so it answers directly instead of escalating
• Spam handling — rules for detecting and ignoring low-quality or automated messages
• Others — anything that doesn't fit the categories above

Scope and Boundaries — teaching the agent your business

Scope and Boundaries is the most important section for getting accurate, on-brand answers. Use it to capture how your business actually works — what you do and don't sell, hours, shipping times, return policy, and how to handle specific scenarios. Phrase each item as behavior: "If the customer asks about X, do Y." The agent generalizes across phrasings on its own, so you don't need to list every wording. Anything covered here is answered directly by the agent instead of being escalated to a human, and it's also where you tell the agent how to interpret data like product custom fields (see the Shopify Integration article for an example).

Escalation tab

Control when and how the agent hands a conversation to a human:

• Escalation Rules — deterministic conditions based on contact attributes and other data that should always trigger an escalation
• Escalation Guidance — fine-tune escalation behavior in specific scenarios that the rules don't capture
• Priority Rules — override how the agent classifies priority on escalation. Without rules it uses sensible defaults (anger → high, FYI → low, everything else → normal); add rules for business-specific signals like "refund mentions are always high"

Knowledge tab

Give the agent source material to answer from: upload documents (PDFs, text files), paste content directly, or point it at your website URL. The agent reads and indexes this content and cites it when answering. See the Knowledge Base Management article for details.

Automations tab

Connect the agent to workflows and integration actions it can run during a conversation — for example, looking up a Shopify order or triggering a published workflow. Which actions are available depends on the integrations connected to your workspace.

Assignment and channels

Decide when the agent takes over a conversation and where it is active:

• Always — the agent handles every new conversation automatically
• When unassigned — the agent only handles conversations with no human agent assigned
• Off — the agent is disabled and will not handle any conversations
• Channels — enable the agent on all channels or only specific ones (for example, the live chat widget and WhatsApp but not email)