A ticket with twenty back-and-forth messages reaches the technician already worn out: before fixing anything, they have to read all of it and still write a clear reply. The NexTool AI Assist module puts AI inside the GLPI ticket itself and, since version 2.x, into the knowledge base as well: it summarizes the history, suggests a reply that cites the article that solves the issue and helps turn what the team solves into an article for the next user.
The problem
In high-volume support, the technician has to read the whole thread before acting: the user's description, colleagues' follow-ups, tasks and automatic replies.
Then comes the writing. Drafting a clear reply takes effort, and an incomplete answer triggers another round of questions that eats into the SLA. There is also a third, less visible cost: the fix found today rarely becomes an article, so the same problem comes back tomorrow as a new ticket.
What AI Assist does
Each feature is turned on separately in the Features tab and granted per profile in the NexTool rights matrix. Together they cover the support cycle:
- Ticket summary - the Summary (AI) button condenses the history into five fixed parts: context, actions, pending items, next steps and relevant notes.
- Suggested reply - a courteous draft with the current status and next steps, signed with the technician's name. When a knowledge base article solves the case, the reply follows its procedure and cites it by title.
- Sentiment and urgency - rates the requester's tone (positive, neutral, negative or critical) and the urgency (low, medium, high or critical), with a 0 to 10 score and a short rationale, shown on the ticket to support prioritization.
- Articles suggested before a ticket is opened - while the user describes the problem in the ticket form (and, in GLPI 11, in native forms), the module suggests knowledge base articles. With a provider configured, it also writes a short answer citing them, and the user chooses between "That solved it, I do not need to open a ticket" and "I will open the ticket anyway". Both choices are recorded.
- Article from a ticket - on a solved or closed ticket, the KB article (AI) button builds a draft with problem, cause, step-by-step solution and verification, created unpublished and linked to the source ticket. If the tasks and the solution do not describe what was done, the module says there is no relevant content and spends no AI credit.
- Assistant in the article editor - suggests three titles and keywords, improves the formatting and, in GLPI 11, writes the short description for the catalog. Everything is previewed and only goes into the article when someone clicks apply.
- Automatic triage and knowledge base gaps - background routines, turned off by default. Triage adds a private follow-up, "Possible path (AI) - validate before applying", and only when an article or a previous solution actually covers the subject. Gap detection flags, every day, categories with many solved tickets and no article, and prepares drafts for review.
The AI answers in the language of whoever is using GLPI. Triage and gap drafts use the system language, because the whole team reads them.
Where each feature shows up
| Feature | Where it appears | Needs an AI provider | After installation |
|---|---|---|---|
| Summary, reply and sentiment | On the ticket itself | Yes | On |
| Suggested articles | Ticket form and GLPI 11 forms, self-service portal included | No (only the written answer does) | On |
| Article from a ticket | Solved or closed ticket | Yes | On |
| Article assistant | Knowledge base editor | Yes | On |
| Automatic triage | Private follow-up, via cron every 2 minutes | Yes | Off |
| Knowledge base gaps | Knowledge base tab, once a day | Yes | Off |
"On" refers to the switch in the Features tab: each action still has to be granted to the profiles that will use it.
How to enable it
- Install the NexTool plugin (base 6.14.0 or later) and go to Setup > NexTool > Modules.
- Enable the AI Assist module.
- In the Connectors tab, choose OpenAI, Google Gemini or Anthropic Claude and paste the API key, which is stored encrypted. The module fetches the provider's model list and preselects the recommended one.
- In the Test tab, use Test connection to validate the key before rolling it out to the team.
- In the Features tab, turn on what the team will use and grant each action per profile in the NexTool matrix.
- For triage and gap detection, set the service user, filters and daily token cap under Settings > Automation. The tasks run through the system cron.
Cost and privacy under control
The module is free; the AI cost is your provider's, billed to your own key. To keep it predictable:
- The Usage tab lists every AI request in a native GLPI grid, with feature, input and output tokens, ticket and user, plus search and export.
- A minimum interval between requests for the same feature, on the same ticket and by the same user, prevents burst clicking (1 minute by default).
- The text sent to the provider has a configurable maximum size (12,000 characters by default), and the summary and reply come from cache until a new follow-up is added.
- Automated routines have their own daily token cap (20,000 by default).
- For OpenAI reasoning models, the reasoning effort is set in the Connectors tab: without that control, a short summary could use up to three times more tokens.
On privacy, the default is conservative: with "Allow sensitive data in the prompt" turned off, the names of the people involved, email addresses, phone numbers and CPF/CNPJ numbers are replaced with role placeholders such as "Requester" and "Technician" before sending. Articles and triage are always masked, because other people will read them. Even so, the ticket content leaves your infrastructure and goes to the chosen provider.
Who it fits (and when NOT to use it)
- High-volume internal help desks, where minutes saved per ticket add up to hours by the end of the day.
- Teams with technicians still developing their written communication.
- Service desks that want fewer repeat tickets and a knowledge base that grows with the support work.
And when should you not use it? With low volume, the gain does not pay for the API: the technician reads the whole thread in seconds. Triage and gap detection depend on well-written articles and solutions; without that material, they simply suggest nothing. And if there are strict rules about sending ticket content to an external provider, stop before enabling it: with sensitive or regulated data, this is a privacy and data protection decision for whoever is responsible for the data, not a technical setting.
What we learned in practice
Two lessons shaped version 2.x. First: AI without a source makes things up. A test ticket with no tasks and no solution once became a draft article with invented steps; today the module first checks whether there is real material, ignores solution template text and courtesy messages, and triage only writes when an article or a previous solution covers the subject and the AI is confident enough. Second: the best ticket is the one that never has to be opened. Suggesting the right article while the user is still describing the problem, and recording when they decide not to open a ticket, shows in the Knowledge base tab how many tickets were avoided and which articles actually help.
Compatibility
- GLPI: 10 and 11 (native forms and catalog description in GLPI 11 only)
- Plugin: NexTool, base 6.14.0 or later
- Module: AI Assist 2.2.3
- Providers: OpenAI, Google Gemini and Anthropic Claude, with your company's own key
- Plan: free (the AI cost is your provider's)
Features, screenshots and a step-by-step guide are on the AI Assist page. The module is part of NexTool, the modular plugin for GLPI. To see it in action or discuss the privacy assessment before enabling it, talk to the team.
Reviewed by the NexTool Solutions team.
