Local LLM Integration

Integrate ServiceDesk Plus with a locally hosted LLM running on Ollama to enable AI-powered features that simplify help desk operations.

Prerequisites

Before configuring the integration, ensure you have the following ready:

• Ollama installed and running on a server accessible to ServiceDesk Plus. Host Ollama on a dedicated system rather than on the ServiceDesk Plus server, to avoid resource contention between model inference and application workloads.

• Infrastructure that meets the minimum hardware and software requirements for your selected model (CPU, GPU, memory, storage, and OS dependencies). Requirements vary by model — check the Ollama model library for the model you plan to run.

• A dedicated GPU for production use. GPU acceleration can reduce response times by three to five times.

• Sufficient disk space and memory provisioned in advance to accommodate the model files and runtime needs.

• ServiceDesk Plus administrator (SDAdmin) access.

Note: ServiceDesk Plus currently supports Ollama’s chat endpoint only. Maintaining sufficient system resources for smooth operation is the customer’s responsibility.

For installation and setup instructions, see Setting up Ollama on Local LLMs. That page also covers the OLLAMA_HOST and OLLAMA_MODELS environment variables and context window sizing.

Enabling Local LLM

To access the Local LLM configuration page:

Navigate to Admin > Apps & Add-ons > Integrations > Third Party.
Under the Local LLM integration card, click Settings.
Select the Enable Local LLM checkbox to activate the integration.

Important: Enabling Local LLM will disable any other LLM integration currently active in the portal.

Configuring the Connection

Once Local LLM is enabled, fill in the connection details as described below.

Referred name

Enter an alias name for your local LLM (for example, “Ollama”). This name is retained even if the integration is later disabled; all other configuration details are removed.

Base URL

Enter the endpoint URL of your Ollama server.

Model name

Enter the exact model name as deployed in Ollama (for example, deepseek-r1 or qwen3).

Enabling AI Features

After configuring the connection details, select the AI features you want to activate. Each feature can be individually enabled or disabled. Use the Permissions control next to each feature to define which users or roles can access it.

Click Save to apply your configuration, or Cancel to discard changes.

Feature Descriptions

The following table describes all available AI features, their functionality, and default permission levels.

Feature	Description	Permission
Approval Prediction	Acts as a fallback when Zia cannot predict a user’s approval action (Approve, Reject, or Need Clarification) from their email reply.	System
Ask Zia (Conversational Chatbot)	Powers the Zia conversational chatbot, using MCP Tools to interpret instructions and perform help desk operations. Requires a model that supports tool calls.	System
Generate Post Incident Review	Generates a post-incident review — summary, root cause, resolution steps, impact assessment, and recommendations — from the request’s subject, resolution, impact description, and last 10 conversations/notes. Best results are achieved with higher-context models.	System
Reopen Prediction	Acts as a fallback when Zia cannot predict whether a user intends to reopen a closed request from their reply.	System
Sentiment Analysis	Analyzes the tone of incoming user conversations for requests and custom modules, categorizing them as Happy, Sad, or Neutral.	System
Zia Agent - Incident Review Summariser	Automatically adds the generated post-incident review as a note to the request.	System
Zia Agent - L1 Support Agent	Suggests approved knowledge base solutions and communicates them to requesters via email.	System
Zia Agent - Request Resolution Agent	Populates the resolution field automatically once a request is resolved or closed.	System
Solution Generator	Generates knowledge-base solutions, including title, content, and keywords — from a simple prompt.	Technician
Ask LLM in Chat	Lets technicians consult the configured LLM in real time from the Actions menu during a requester chat, for AI-suggested replies.	Technician
Classic Bot - Ask LLM	Lets users ask the configured LLM questions directly within the Zia bot.	All (configurable)
Classic Bot - Context-aware Solution Suggestion	Uses a RAG approach to surface relevant solution summaries, with a reference link, when a user reports an issue, requests a resource, or searches for answers in the Zia bot.	All (configurable)
Classic Bot - Explore solutions	Searches the web for external solutions and generates answers within the Zia bot when no internal match is found.	All (configurable)
Classic Bot - Explore solutions while creating request	Surfaces external solutions from the web while a user is creating a request through the Zia bot.	All (configurable)
Classic Bot - Template Prediction	Predicts the most relevant request template from the user’s description in the Zia bot. Advanced models provide more accurate predictions.	All (configurable)
Reply Assistant	Generates content, drafts replies, rephrases text, and checks grammar in the reply/forward editor.	All (configurable)
Summarization	Generates a summary of request details from the description and the last 10 conversations/notes, for both request and custom modules.	All (configurable)
Writing Assistant	Creates and refines content in editor fields, including emails.	All (configurable)

Managing Permissions

Each AI feature has a permissions control that determines who can access it:

• System-level features (such as Approval Prediction and Reopen Prediction) run automatically in the background and are not configurable by role.

• Role-based features (such as Reply Assistant and Summarization) can be restricted to technicians or made available to all users (technicians and requesters) using the dropdown next to each feature.

Monitoring Usage and History

Two controls are available at the top right of the Local LLM Configurations page:

• Usage Stats – Monitor Local LLM token consumption and API call metrics to keep track of costs and usage trends.

• View History – Review a log of all configuration changes made to the Local LLM integration, including timestamps and the users who made them.

Performance

Self-hosted models ran eight to 10 times slower than cloud-hosted models in testing. A dedicated GPU is strongly recommended for production use.

To handle queries from multiple users at the same time, enable parallel request processing. See the Ollama FAQ for details.