Skip to main content

Create agents with Auto mode

note

Beta: This feature is currently in beta and may have limited availability. Functionality is subject to change based on ongoing improvements and feedback.

Auto mode is a goal-driven way to build agents: you describe what the agent should accomplish in natural language, choose which tools and data sources it may use, and Glean plans and runs the steps to get there. You do not have to wire every branch or maintain a rigid workflow diagram for each scenario.

Use Auto mode when you want to move quickly without deep technical setup, or when the right order of steps depends on what the agent learns as it works.

In Auto mode, you grant access by app, for example Jira, Slack, Gmail, or Google Docs rather than selecting individual actions. Glean uses each allowed app to reach the capabilities it needs, while the existing authentication, approvals, and runtime guardrails of your organization still apply. Keep the app list as small as practical—only what the agent needs for the job—so runs stay reliable and focused.

The app-level model applies to Plan & Execute and Auto mode agents. Static workflow agents still use explicitly selected steps and actions.

Auto mode compared to Workflow mode

You can build agents in two ways.

Auto mode

You begin by describing the agent's purpose in natural language. Glean reasons, plans and executes the steps and actions automatically. Auto mode is best when flexibility is important or when you prefer not to build structured flows.

Workflow mode

You define each step in the visual Agent Builder. Workflow mode is best when you require deterministic logic, complex branching or strict control of execution.

Requirements and Access

Auto mode is a beta feature. Your Glean environment must be on the latest Agentic Engine and must have Agents enabled. Contact your account team to activate this capability.

Create an Auto mode agent

  1. Open the agent library
  2. Select Agents
  3. Select Create agent
  4. Choose Auto mode
  5. Select either Assist Me or Setup

Auto mode options

You can switch between the two at any time. Both paths produce the same type of agent.

Build with Assist Me

Assist Me provides a conversational interface. When you use this mode we recommend that you describe:

  • Goal and role
  • Scope and data
  • Allowed tools and actions
  • Expected output format
  • Example tasks

As you provide details, auto mode generates and updates the instructions of the agent, knowledge sources and allowed action packs. A preview panel lets you test your agent throughout the process.

Build with Setup

Setup provides a structured editor with three key sections:

  • Instructions
  • Knowledge
  • Actions

This mode is useful when you want to edit configuration fields directly. You can switch back to Assist Me whenever you want to iterate conversationally.

Attach Knowledge

Connect the files, folders, collections or data sources the agent should rely on. Auto mode may propose knowledge sources, although you choose which ones to include.

Add conversation starters

Conversation starters appear when a user opens your agent. They give quick examples that help users understand what the agent can do and offer easy starting points for most valuable or common tasks that the agent can help with.

Select apps

Choose only the apps required for the agent to complete its job. For example, include Confluence app only if the agent responsibilities are mainly around Confluence. Limiting actions improves reliability.

Save and share

After configuring and testing your agent:

  1. Give it a clear name.
  2. Select Save to apply your changes and expose the latest version to your intended audience.
  3. Share it with your team.
  4. Review usage in Agent Insights.

Best practices

  • Reference authoritative sources whenever possible.
  • Test in both Fast mode and Thinking mode if available and compare performance.
  • Validate the agent with multiple types of tasks before saving.

FAQs