GLPI AssistIA

Use Case

Incident management, especially in IT departments, requires tasks that are often repetitive, such as reviewing wikis, documentation, or executing verification commands on specific servers. GLPI AssistIA aims to serve as a foundation for reducing all this workload by providing a preliminary report on the incident to be handled. Once an incident is registered in the system, it is sent to a server (GLPI AssistIA Server) which searches for all related information and generates a report with possible solutions. This report will be visible to the agent handling the incident as a private note at the moment they take the case, so they only have to worry about providing the best possible service to the client.

Evolution of the Agents

The GLPI AssistIA agents have evolved through several prototypes, each one adding new functionalities and capabilities to improve incident management. Next, the main milestones in this evolution are described.

Prototype 1

A basic agent was implemented with a single function to summarize a GLPI ticket, sending the result to a file on the server.

Prototype 2

The ability to perform a ping to an IP address provided in the ticket was introduced, and the reading of Wiki.js documentation was improved. This allowed the agent to respond more effectively to connectivity incidents.

Prototype 3

Multiple agents with specific roles (Analyst, Resolver) were integrated, and a workflow was implemented to allow agents to interact and collaborate on resolving more complex problems. The concept of an MCP data bus was introduced.

LLM Model Analysis

For the development of GLPI AssistIA, a comparative analysis of several language models was conducted to determine which one would be most suitable for the project. The goal was to find a model that offered a balance between processing capabilities, cost, and the ability to be self-hosted to ensure data privacy.

The orchestration of the agents was carried out using the CrewAI library, which facilitates collaboration between different specialized agents. Both open-source and proprietary models were considered, with special attention paid to their flexibility for integration into the server. This analysis was crucial for optimizing the system's performance and efficiency in the GLPI environment.

For more details on this analysis, you can consult the full document in the project's repository.

See the full analysis document (Spanish)

Architecture and Overview

The core of the system is designed around a workflow that is activated when a ticket is created in GLPI. The ticket information is processed by a system of intelligent agents (CrewAI) that collaborate to analyze, enrich, and propose solutions. This CrewAI connects to external tools (knowledge bases, monitoring systems) through a MCP (Model Context Protocol) Server, which acts as a data bus.

Ticket Workflow

GLPI Entry: A user or technician creates an incident ticket.
Agent Activation: The incident is transferred to the CrewAI system hosted on GLPI AssistIA Server for processing.
AI Agent Analysis: A team of agents evaluates the urgency, classifies the incident, and reviews historical data to propose a solution.
GLPI Response: The generated solution and analysis are published in the GLPI ticket, assisting the technician or responding directly to the user.

Configuration and Usage Example

Key Features

Ticket Summary and Enrichment: The AI analyzes and summarizes the user's problem, adding technical context.
MCP Architecture: A decoupled data bus to facilitate communication and scalability.
Intelligent Contextual Information: Provides relevant information to both technicians and users.

Practical Cases

Plugin Interface

Success Metrics

The project’s success will be measured by achieving the following objectives:

Reduction of over 70% in first response time.
Accuracy of over 85% in automatically generated responses.
Reduction of over 50% in tickets that need to be manually escalated.
Reduction of over 40% in average incident resolution time.
User satisfaction level above 4.0/5.0.

Collaboration

This project was made possible thanks to the ANFAIA Summer Grants program and the collaboration of Aitire.