Decide with AI
Have AI sort text into options you name, such as spam, toxic or fine.
integration.decide reads some text, asks a question about it, and takes the way out of the option that fits best. Use it to moderate messages, route support questions or sort replies.
Settings
| Setting | What it does |
|---|---|
| What to judge | The text to read, such as {message.content}. Up to 8,000 characters are read. |
| Question | What to decide, such as “Which best describes this message?” |
| Options | One per line, as Name: what it means. Each gets its own way out. Saying what an option means helps it pick well. |
| Model | Fast suits most checks. Careful is a larger model that reads more closely. |
Ways out
One for each option, named as you wrote it. Turn on If it fails for a red way out, taken when AI can't be reached or the bot has asked too many times this minute.
Produces
| Token | Value |
|---|---|
{ai.choice} | The option picked, as you named it |
{ai.confidence} | How sure it was, from 0 to 1 |
{ai.NAME} | Each option's chance, from 0 to 1, with its name written in lowercase with dashes |
Notes
- A bot can ask up to 60 times a minute. Past that, the step fails until the minute is up.
- A test run doesn't ask AI. It goes down the way you choose, or the first option.
- Reading a message's text in an event flow needs the Message Content intent.
Example
When a message is sent, decide between Fine, Spam and Toxic. From Spam, delete the message. From Toxic, delete it and time the member out for 10 minutes. Check {ai.confidence} with a condition first if you only want to act when it's sure.