Skip to content
WeZend

Automation

Chatbot actions: lookups and approved requests

Let the chatbot look things up in your own system and ask for actions that only a signed-in admin can approve, so nothing changes without a person.

Answers from text only go so far. With actions, the bot can look up a live value in your own system, such as "where is order 12?", or ask for something to be done. You define actions on the Chatbot page, and only owners and account admins can change them. Start with AI chatbot if you have not set the bot up yet.

Read-only lookups

A lookup is a read. You define:

  • A name and a description, so the bot knows when the lookup is useful.
  • An https address template with placeholders, such as a path with an order number in it.
  • Up to four typed values the bot may fill in. Each value is checked for type, length and allowed characters before it is used.
  • An optional secret header, for example an API key. It is stored encrypted and never shown again, never put in a prompt and never logged.
  • An allow-list of returned fields. Only the fields you list ever reach the model.

The bot can only ask for a lookup by name with values. The host is fixed by your template, public hosts only, redirects are never followed, and slow or oversized answers are cut off.

Taking a value from the sender

A value can be taken from the verified sender instead of from what they typed, for example their phone number on SMS or WhatsApp. The lookup then only ever answers about the person who is writing.

Safe to share

The assistant repeats what your system returns to whoever is chatting. Treat every lookup accordingly:

  • A lookup by an order number or other guessable id is open to anyone who can guess one.
  • Prefer values taken from the verified sender, and combine them with a field allow-list.
  • Return status and tracking, not addresses, phone numbers or e-mail addresses.
  • Test with the test console, which never calls your system for real.

Actions with an effect

An effect action uses POST, PUT, PATCH or DELETE, with an optional JSON body template. The bot cannot carry one out. It can only request it:

  1. The bot files a request, and the contact is handed to a person.
  2. Your people are notified by e-mail.
  3. A signed-in admin or owner sees the method, the address, the exact body and the values, and approves or declines. Agents, members and API keys cannot approve.
  4. One approval is exactly one call.

If the outcome is unknown, for example after a timeout or a server error, WeZend says so plainly and never retries; check in your own system. A request expires after 24 hours, and if you edit the action after the request was made, approving it sends nothing. There are limits on how many requests can wait per conversation and per account.

Give effect actions the narrowest possible scope, and approve with the same care as if you had typed the request yourself.

One platform. Every customer interaction.

Replace your patchwork of messaging APIs, CDP and automation tools with a single engagement platform built for scale.

No credit card required · EU data residency · 99.99% uptime SLA