User Guide · AI Agents
Actions
Actions let the agent do things, not just talk: close chats, escalate, tag, update contacts, show products and create orders.
Actions are the tasks an agent can perform on its own during a conversation. You control them in the agent editor's Actions tab. No setup is needed: switch an action on and the agent starts using it when appropriate.
- 1Close conversations when the customer confirms the issue is resolved.
- 2Smart escalation to human.
- 3Fields to auto-collect: Name, Email, Phone, Country, Company, Language.
- 4Auto-tag conversations.
Core actions
| Action | What the agent does |
|---|---|
| Close conversations | Closes the conversation when the customer confirms their issue is resolved. |
| Smart escalation to human | Hands the chat to your team when the AI cannot help or the customer is upset, with a priority (low, medium, high or urgent). |
| Update Lifecycle stages | Moves the contact between your lifecycle stages based on buying signals. |
| Auto-collect contact info | Saves details the customer mentions, such as name, email, phone and country, to the contact. |
| Auto-tag conversations | Adds relevant tags, reusing your existing tags or creating new ones. |
| Add internal notes | Leaves notes for your team inside the conversation. |
Rich messages
Independently of the switches, the agent can format its replies for the channel:
- Send images.
- Quick-reply buttons (2-3 options) and lists (4-10 items).
- A single link button, for example “Track your order”.
- React to a message with an emoji or reply to a specific message.
- Update custom contact properties.
- Show your past work from the portfolio library as a carousel or a pop-up.
Store actions
In Store Connection, switch on Link to Store and pick a store. The agent gains:
For Shopify and Salla stores the agent can also add to and remove from a cart and send the checkout link. See Shopify and Salla.
Bookings
If you set the agent's Goal to Booking / Scheduling, it asks which service the customer wants, collects the details, records them as an internal note and escalates complex requests to your team, who confirm the appointment.
Custom actions (your own API)
Developers can register Remote Actions: read-only lookups in your own system, for example “get my subscription status”. An agent uses a remote action only when that action's name appears in the agent's instructions, so you decide exactly which agent can call what.
Seeing actions in tests
In the test chat, each reply shows chips for the actions the agent would take, so you can confirm it escalates, tags and orders when you expect it to, without touching live data.