​
Overview

Glean’s Wait for user input action allows agents to pause execution at any point to collect information or confirmation directly from a user. This makes it possible to build workflows that incorporate real-time human interaction, unlocking new use cases that require critical gating or decision points. This feature is designed for agent creators and users who need their automated workflows to depend on live or asynchronous human input. For now, it is only available to run manually in the Glean app. Agents using this action cannot be scheduled, published to Slack, used through MCP, or Glean’s APIs.

​
Usage Example

The Wait for user input action is invaluable in workflows where automation must be stopped for human input, such as:
  • A time off request agent that asks users for the dates they plan to be absent
  • A sales coaching agent that extracts the name of an account executive from a transcript and asks the user to confirm that it has made the correct choice

​
Configuration

The Wait for user input step can be inserted into any part of the agent graph. When configuring the wait for user input step, you can set a natural language goal (what you want to ask the user) and, optionally, an exit condition (when execution should abandon/skip waiting).

​
How It Works

​
Agent Builder Experience

Use the Agent Builder to add a Wait for user input step anywhere in your flow and define exactly what the agent needs from the user.
  1. Insert a Wait for Input step: Add it at the point where downstream steps depend on user input or a decision.
  2. Write the goal: Describe in natural language what the agent should ask a user for, including constraints and examples the agent can validate against. 
    Prompt the user to enter their PTO start date and end date. Validate the following conditions:    

- The start date must be after today's date.
- The end date must be after today's date.
- The start date must be earlier than the end date.
  3. Configure input expectations: Specify the fields or information the agent must collect (e.g., date ranges, IDs, or approvals). The agent may ask clarifying questions until the inputs are valid. 
    When requesting or interpreting date inputs, ensure both dates explicitly include the day, month, and year, formatted as "YYYY-MM-DD" (e.g., "2025-09-10"). If a user refers to a date using relative qualifiers (such as "next Friday" or "the week of Thanksgiving"), convert these references into specific dates and clearly present them for user confirmation. For example, respond with:  "Did you mean '2025-09-12' when you said 'next Friday'?" Always ask the user to confirm the interpreted date to avoid ambiguity.`
  4. Connect downstream usage: Map collected values to later steps (e.g., filling tool parameters, branching decisions, or guards). You may reference the input fields collected using a similar natural language you used in the Wait for user input step. 
    Downstream “Workday: Request Time Off” step Instructions: File a PTO request based on start date and end date.
  5. Test your agent: Run the agent in preview mode to ensure it works as intended, providing clear interactions that validate correctly.
For the Wait for user input step and the following steps, we recommend keeping Manage memory in Advanced Settings to All previous steps (the default option). This allows each step to reference context from earlier steps and ensures all collected inputs are retained.

​
End User Experience

  • When the flow reaches the Wait for user input step, execution pauses and the user is prompted for input (message or form), with follow-up clarifications as needed.
  • Agents cannot be cancelled while they are waiting for user input.
  • Conversation is continuous—all prompts and responses remain in the same chat thread for context.

​
FAQ