Agent Management

Guide for creating, configuring, and utilizing agents in AI Hub

Key Features

AI Hub agents go beyond simple chat functionality to serve as business-specific AI interfaces that connect various tools and knowledge within an organization. Users input requests in natural language, and agents provide automated responses based on defined purposes and tools.

Concept Definition

Agents are composed of the following elements:

• Instruction (Action Guidelines): Defines what roles and actions the LLM should perform
• LLM Model: Selects AI models like GPT-4, Claude to provide the intelligence foundation for agents
• MCP Tool (Interface with External Systems): Integrates with external systems like GitHub, Notion, Snowflake
• Knowledge Bundle (Vector-based Organizational Knowledge): Vectorizes organizational documents and data to utilize as searchable knowledge

This combination is configured in the admin interface, and users can utilize agents shared within the organization.

Agent Components and Operating Principles

AI Hub agents are defined by the following components.

Agent Types

Agents are classified into two types:

• Built-in: Agents that can be installed through the agent library and used immediately. These can be provided by QueryPie AI Hub or pre-configured for specific organizational needs.
• Custom: Agents that organizational administrators can create by directly configuring agent components within AI Hub.

Agent Status

Agents have one of the following three statuses:

• Draft: Status immediately after Custom Agent creation or Built-in Agent installation. Users cannot use the agent in this state.
• Enabled: Status where configuration is completed normally and the agent can be shared and executed by users. Once changed to Enabled, it cannot return to Draft status, and modification and deletion of the agent are not possible in Enabled status.
• Disabled: Temporarily deactivated status where the agent is not exposed to users.

Agent Management Features

Custom Agent Creation

Click the “Create Agent” button to input agent name and summary information, then create a Custom agent.

Administrators can configure the following items in the agent detail screen:

• Agent Name: Set a unique name for the agent. It’s recommended to use a clear name that’s easy to identify within the organization.
• Summary: A short description summarizing the agent’s functionality in one line. Users can refer to this when selecting agents.
• Icon: Set an icon to visually distinguish the agent.
• Description: Detailed description of the agent to be displayed to users. Clearly explains the tasks the agent can perform and its main features.
• Prompt Guide: Example questions to guide users. Up to 5 can be registered, showing users how to utilize the agent.
• LLM Model: Select the AI model the agent will use. Currently provides Claude 4 Sonnet model as the default.
• Instruction: Define the agent’s role and action guidelines. Specifies how the LLM should respond, what tone and style to maintain, etc.
• MCP Tool Addition: Add tools that can integrate with external systems. Can connect with various services like GitHub, Notion, databases.
• Knowledge Bundle: Connect knowledge sources that the agent can reference. Can utilize organizational documents, databases, files as vectorized searchable knowledge.
• Test Chat: Immediately test the agent’s behavior via text chat on the right side of the screen (Note: File uploads and the Artifact feature are unavailable).

MCP Tools can be added in two ways:

• Select MCP servers and tools provided by AI Hub
• Direct registration of Custom Remote MCP

Agent Detail Page

The agent detail page provides different management features depending on Custom Agent and Built-in Agent.

Custom Agent

An agent created directly by administrators with complete freedom to modify all components.

• Full Edit Permissions: Can modify all settings including agent name, description, Prompt Guide, Icon, LLM model, Instruction
• Tool and Knowledge Management: Free configuration including MCP Tool addition/deletion, Knowledge Bundle connection/disconnection
• Advanced Settings: Can adjust detailed options like Artifact activation, Max Token settings
• Immediate Save: Changes are reflected immediately, allowing real-time agent behavior verification

Built-in Agent

An agent installed from the agent library with limited edit permissions.

• Limited Edit Permissions: Only agent name can be modified, core settings cannot be changed
• Tool Integration Only: Only MCP Tool authentication and integration settings can be added
• Fixed Configuration: LLM model, Instruction, Knowledge Bundle, etc. maintain pre-defined states
• Stability Priority: Core setting changes are restricted to maintain original agent stability

Share Scope Settings

An item for setting agent sharing scope with the following two sharing ranges:

• All Users: All members of the organization can access this agent
• Specific Users: Select specific users by searching for them by name or email to allow access to the agent.

Agent Library

The agent library is a feature that allows organizations to install and use pre-built Built-in agents provided by AI Hub.

Key Features

• Pre-configured Agents: Provides agents optimized for specific tasks or domains
• Immediately Usable: All configurations are completed for immediate use after installation

Installation and Usage

1. Agent Selection: Select desired agent from the library
2. Installation: Click “Install” button to install agent in the organization
3. Configuration: Configure MCP Tool integration and authentication as needed
4. Activation: Activate agent to share with users

Agent Usage in User Interface

When administrators set tool authentication methods belonging to agents to “Require User Authentication”, users need to directly input configuration information or authenticate in the Configuration tab of the detail page

For detailed user agent usage guide, refer to User Guide > Agents.