Skip to content

AI Agent#

The AI Agent page is used to manage the logical AI agents that can be used by AI interactions in Critical Manufacturing MES.

An AI Agent stores the canonical definition of an agent inside MES, independently of the external runtime platform. This allows the system to manage both system-defined and user-defined agents in a consistent way.

Overview#

Use an AI Agent to define:

  • The display information shown to users.
  • The instructions that describe how the agent should behave.
  • The MCP Servers that the agent is allowed to use.

Main View#

Depending on the configuration, the main view can include the following sections:

  • Details - displays the main information of the AI Agent, such as the name, description, display name, and additional information.

    AI Agent Details section.

  • Instructions - displays the AI Agent behavioural prompt.

    AI Agent Instructions tab.

  • MCP Servers - displays the MCP Servers associated with the AI Agent and their selected tools.

    AI Agent MCP Servers section.

  • Attachments - shows the documents that are attached to the AI Agent and enables you to attach additional documentation to it.

  • History - displays previous changes to the AI Agent, when history tracking is available.

Manage Instructions#

Instructions define how an AI Agent should behave when it processes a prompt.

Supported Formatting

The system retains only the formatting supported by the Instructions editor during synchronization with LibreChat.

To view or update instructions for a user-defined AI Agent:

  1. Open the AI Agent.
  2. Select Instructions on the Details View.
  3. Select Edit.
  4. Enter or update the instructions.
  5. Save the AI Agent.

Instructions Wizard to Manage Agent Instructions

Instructions for system-defined agents are read-only. Use a user-defined AI Agent when scenario-specific instructions are required.

Select a Provider and Model#

When you create or edit a user-defined AI Agent, select the Model that the agent uses. The Provider is defined by the Generative AI deployment and is displayed as read-only when only one provider is available.

For a system-defined AI Agent, administrators can update the Model only. The Provider remains read-only.

The available models depend on the models configured for the environment. Contact your system administrator if the required model is not available.

Manage MCP Servers#

Use Manage MCP Servers on an AI Agent to select the MCP Servers that the agent can use.

  1. Open the AI Agent.
  2. Select MCP Servers on the Details View.
  3. Select Manage.
  4. Add or remove the required MCP Servers.
  5. Save the changes.

The selected MCP Servers determine which tools are available to the AI Agent in the runtime. The system synchronizes the updated association with LibreChat.

MCP Servers Wizard to Manage MCP Servers

Terminate and Restore an Agent#

You can terminate a user-defined AI Agent from the page actions. System-defined agents cannot be terminated.

Terminating an agent removes it from LibreChat and terminates its associations with MCP Servers. It does not terminate the associated MCP Servers.

To restore a terminated agent, select Unterminate. If the agent is configured to publish to the runtime, the system synchronizes it with LibreChat. Before restoring the agent, ensure that all MCP Servers previously associated with it are active. After restoring the agent, use Manage MCP Servers to add the required associations again.

Configuration Notes#

Keep the following behavior in mind when configuring an AI Agent:

  • Only administrators can create or update AI Agents.
  • For system-defined agents, only the Model is editable.
  • User-defined agents can be adjusted to support scenario-specific instructions and MCP Server associations.
  • For the LibreChat runtime, an optional Display Name can contain only letters and spaces.

Synchronization with LibreChat#

The system synchronizes AI Agents and their associated MCP Servers with LibreChat, the current AI runtime. For more information about runtime availability and configuration, see LibreChat Integration.