AI Actions Overview
AI Actions let your agent automatically trigger real-world workflows mid-conversation — sending emails, calling APIs, sharing booking links, or handing the conversation to a human — without you lifting a finger. Each action is scoped to a single agent and fires based on trigger keywords you define.
Four action types
| Type | What it does | Best for |
|---|---|---|
| Sends an HTML email to a specified address with conversation data in the subject and body | Lead notifications, support alerts, internal escalations | |
| API / Webhook | Makes an HTTP request (GET, POST, PUT, PATCH) to any URL with a JSON body you define | CRM updates, ticket creation, Slack alerts, custom integrations |
| Calendar | Returns a Calendly or Cal.com booking link in the chat so the visitor can self-schedule | Demo bookings, support calls, consultation requests |
| Human Handoff | Transitions the conversation into the human handoff flow — the agent stops answering and your team takes over | Escalations, sensitive requests, anything the AI shouldn't handle alone |
How triggering works
Each action has a Trigger keywords field (comma-separated phrases) and a When to use description. The system checks the visitor's latest message against your trigger phrases in two ways:
- •Single-word triggers — matched as a whole word using word-boundary rules (e.g. "book" won't match "Facebook").
- •Multi-word triggers — matched as a substring anywhere in the message (e.g. "schedule demo" matches "I'd like to schedule a demo call").
- •The AI also uses the "When to use" description to intelligently decide when to fire the action, even without an exact keyword match.
- •Trigger phrases must be at least 3 characters. Matching is case-insensitive.
Execution tracking
Every time an action fires, the system records an execution entry with:
- •Status — pending, success, or failed.
- •Payload — the data that was sent.
- •Response — the server response (HTTP body, truncated).
- •Error message — populated on failure.
- •Timestamp — when it fired.
The action card in your dashboard shows the total execution_count and last_executed_at so you can see how active each action is at a glance.
Human Handoff actions
A Human Handoff action tells the agent to stop answering and hand the conversation to your team — triggered either by a keyword match (e.g. "talk to a human", "speak to an agent") or by the AI itself deciding the conversation needs a person, based on the "When to use" description you write. When it fires, the conversation moves into your Human Handoff flow, using whatever escalation rules and destinations you've configured there. If the visitor hasn't shared contact details yet, they're shown the lead capture form first so your team has a way to follow up.
AI Actions vs Zapier
| Feature | AI Actions | Zapier Integration |
|---|---|---|
| Scope | Per agent | Organization-wide (all agents) |
| Trigger | Conversation keyword / AI intent detection | System events (lead captured, conversation started/ended, message received) |
| Setup | Settings → AI Actions tab of each agent | Settings → Integrations → Zapier |
| Use case | Mid-conversation workflows (email alerts, booking links, CRM webhooks) | Post-event automation (add to Google Sheets, notify Slack on new lead) |
Plan limits
The number of AI Actions you can create per agent is determined by your plan. Your current limit is shown at the top of the AI Actions tab. Upgrade your plan to increase the limit.
Creating your first action
Open your agent
Click AI Actions
Choose an action type
Fill in the form
Add trigger keywords (optional)
book demo, schedule call, set up meeting.Save and test
Was this page helpful?