Custom Actions
Let the assistant call your API or show a link of yours, with no code.
Give the assistant actions of your own — with no code. You set them up in Dashboard → Integrations, on the Custom Actions tile. How that screen works is covered in Integrations.
What it does
You describe when the assistant should use an action, in a plain sentence, and it decides during each conversation whether to. An action does one of two things:
- Call my API — the assistant calls your HTTPS address during the chat and reads the answer back to the visitor: look up an order, check availability, fetch a quote. It can collect input through a small form first. The call is made from our servers, and only to public
https://addresses. - Show a link — the assistant shows a button or link — a tracking page, a booking page, a portal — and can drop the visitor's answers into the address. No server is called.
Each active action becomes a tool the assistant can use. To simply notify your systems that something happened, use Webhooks instead; custom actions are for the assistant fetching an answer mid-conversation.
What you need
- For Call my API: a public
https://address that answers the request, and the auth headers it needs to accept it. - For Show a link: the address of the page to open.
Which plans include custom actions, and how many can be active at once: see Plan and limits.
How to connect
- Open the Custom Actions tile and click New action.
- Pick a starting point from the gallery — Check via API, Show button, or Custom API to start from scratch.
- Fill in the action (see Settings) and click Save action. Save action and Cancel sit at the start of the action, under its title. If you have changed something, Cancel or Back asks Discard your changes? before it throws your edits away. Save action stays off until the required fields are valid; a message under a field says what to change.
The editor is roomy; on a wide screen the side panel opens wide to give it space, Compact narrows the panel and Wide widens it again, and this browser remembers your choice.
Settings
- Name — an internal label, which also names the tool. Visitors do not see it.
- When to use — the trigger. The assistant reads it to decide when to run the action, so write it the way you would explain it to a teammate: "When the visitor asks where their order is."
- What the action does — Call my API or Show a link.
- How the visitor sees it:
- Just text (silent) — the action runs quietly and the assistant tells the visitor the result in chat;
- A form — the visitor fills in the fields first, then the action runs;
- A card / link — the result appears as a card or a button.
- Fields the assistant collects — each with a key, a label, a type (Text, Email, Phone, Long text, Options to choose from, Checkbox) and whether it is required. For Options to choose from, list the options in the box beside it, separated by commas.
- Active — the switch that lets the assistant use the action. Switch it off to keep the action without it being offered.
Call my API
- URL — the public
https://address the assistant calls; the collected fields are sent to it. - Method, and Send fields as a JSON body, a Form-encoded body or a Query string.
- Auth headers (optional) — stored encrypted and never shown again. When you edit the action, leave them empty to keep the stored values while the URL stays on the same host, port and scheme; a URL on another host, port or scheme clears them. The editor shows whether headers are saved. Header names that belong to us are not accepted:
Content-Typeand the threeX-NovixAgent-…headers that signed requests carry (see Webhooks). Saving an action that sets one is refused; use a different header name.
Show a link
- Link template — the address to open, with
{{ key }}placeholders, where key is the key of a field the assistant collects; the visitor's answer replaces it, and a placeholder with no answer is left empty. The button label takes the same placeholders. - Button label — the text on the button.
Check that it works
In the editor, click Test preview to see roughly what the visitor will see. Then open your website's chat and ask something your When to use sentence covers, and check that the assistant runs the action.
If it isn't working
| What you see | What to do |
|---|---|
| The assistant doesn't run the action when you expect it to | Check that the action is active (an inactive one is marked off), and make When to use describe the visitor's request more plainly. |
| Plan limit reached — upgrade for more | Your plan's number of active actions is used up. Switch one off or delete it, or move to a larger plan. |
| Could not save action | A message under a field says what to change; fix it and save again. With no message, try again; the change was not saved. |
| Could not load actions | Try again. |
| A Call my API action gives the visitor no result | Your address may have refused the call or not answered. Check the URL, the method, the auth headers, and that the address is public. |
Disconnect, and what happens to your data
Switch an action's Active switch off, in its editor, to keep it without letting the assistant use it. Delete, in the list of actions, removes it for good after you confirm — the assistant stops offering it, and it cannot be restored; you can only create it again.
Conversations in which the action already ran are not changed.
Plan and limits
How many actions each plan keeps active at once: What each integration allows, by plan. The panel shows how many of yours are in use.