Skip to main content
You can create an agent using natural language, allowing you to describe what you want in plain language and have the Agent Builder automatically generate the appropriate steps. This approach provides a quick first draft that you can further refine using either natural language instructions or manual editing.

​
Start creating your agent

  1. Navigate to the Agent Library.
  2. Select the Create agent button.
  3. In the modal that appears, select Build with Natural Language.
  4. Enter your instructions in the natural language panel:
    • Describe the steps this agent should follow to complete this task as well as how the final output should be generated.
    • If you would like the agent to use the knowledge from a specific document that is indexed by Glean, you can either enter the document’s URL or tag it using ’@’.
    • You can specify input fields by enclosing them in double brackets (e.g. “Summarize the [[Zendesk ticket]]”).
  5. The Agent Builder will automatically generate your agent’s steps based on your instructions.

​
Example instructions

Create a task based agent that asks the user to enter a Jira ticket and then suggests how to resolve the ticket using the contents of relevant playbooks: 
Suggest how to resolve the eng oncall issue in [[Jira ticket]]
* Search for playbooks that are related to the content of the ticket
* Think about which playbook is most relevant to the ticket
* Read the full content of that playbook and summarize how to resolve the ticket
Create a conversational agent that answers questions using past user research studies: 
Create a chatbot that answers questions about user research studies 
using these documents:
<URL of user research report 1>
<URL of user research report 2>
<URL of user research report 3>

​
Refining your agent

After the initial draft is generated, you have multiple options for refining your agent:
  • Continue with natural language: Click “Open Panel” in the upper left corner to access the natural language panel and provide additional instructions to modify your agent’s steps.
  • Manual editing: Switch to manual editing at any time to make precise changes to individual steps, models, triggers, or configurations.
  • Enhance prompts: Use the Enhance prompt feature on individual steps to apply best practices and improve instruction quality.
You can freely switch between these approaches throughout the building process, using whichever method is most efficient for the changes you want to make.