How do Natural Language Triggers work?

When you configure a natural language trigger, it makes it possible for the Moveworks Copilot to present your plugin to users when they ask for it.

Our natural language triggers are built based on a full understanding of the plugin's contents. During configuration, the Agentic Reasoning Engine analyzes the plugin and confirms its understanding with the developer through a contrastive learning experience.

Then, when the user asks your Moveworks Copilot a question, the Manifest Generator takes the following steps:

1. Uses the natural language trigger to filter to the best plugins
2. Uses the plugin title, description, details, & trigger examples to rewrite model-facing plugin descriptions
3. Optimizes the end-user experience with the Agentic Reasoning Engine.

How to write good triggering examples

You should:

• Add triggering examples that span a few diverse examples.
• Never add the same / semantically identical triggering examples as both positive or negative examples for the same plugin.
• Do not add positive examples for vague or generic utterances such as "ticket" or "comment".

Additionally, depending on your situation you should either add positive or negative triggering examples.

Triggering Instructions

Along with triggering examples, you can add triggering instructions to your plugin descriptions. Here are examples:

• Clarify when to use the plugin (e.g. You can only search and query meetings, you cannot do any other read, update, or write actions such as cancel, edit, update, delete.)
• Clarify when to not use the plugin (e.g. Do NOT use this plugin for checking common availability between multiple users, use the "find_availability_between_users" plugin for that purpose. DO NOT attempt to look up someone else's calendar.)

Troubleshooting

My plugin is not triggering even when I type the same thing

Please note that plugins may take up to 2 minutes to update in your AI agent. Please try again in a few minutes.