8.1 Connecting to LLM
Most AI features in IDMP — panel generation, analysis suggestions, AI Chat, root cause analysis — require a connection to an external Large Language Model (LLM). IDMP uses an OpenAI-compatible interface, so any LLM provider or self-hosted model that exposes an OpenAI-compatible API can be used.
Built-In Trial Connection
IDMP ships with a built-in trial AI connection that is active for 15 days after installation. During the trial period, all LLM-dependent AI features work immediately without any configuration. Once the trial expires, you must configure your own AI connection to continue using these features.
TDgpt-based features (anomaly detection, forecasting, missing data imputation) are independent of the LLM connection. They require the TDgpt module to be installed alongside IDMP.
Configuring an AI Connection
AI connections are managed in the Connection Management section of the system settings, alongside TDengine data connections.
To add or edit an AI connection:
- Navigate to Settings → Connection Management.
- Click + Add Connection and select the AI connection type.
- Fill in the connection fields:
| Field | Description |
|---|---|
| Connection Name | A unique name to identify this AI connection |
| API Endpoint | The base URL of the OpenAI-compatible API (e.g., https://api.openai.com/v1) |
| API Key | The authentication key for the API. Leave blank for local deployments that do not require authentication. |
| Q&A Model | The model used for standard natural language queries and panel/analysis generation (e.g., gpt-4o) |
| Deep Thinking Model | The model used for complex analytical tasks that require extended reasoning, such as root cause analysis (e.g., o1 or o3) |
- Click Test Connection to verify the endpoint and credentials.
- Click Save.
Two Model Configuration
IDMP uses two separate models from the same AI connection:
- Q&A Model — handles everyday interactions: answering natural language queries, generating panel suggestions, creating analysis configurations, and narrating panel insights. This model should be fast and cost-effective.
- Deep Thinking Model — handles computationally intensive tasks that benefit from extended reasoning chains, most notably Root Cause Analysis. This model can be slower and more expensive; it is only invoked when deep analysis is explicitly requested.
In the AI Chat interface, users can toggle Deep Thinking mode to route their query to the Deep Thinking Model instead of the Q&A Model.
Local Deployment
For organizations running a self-hosted LLM (such as a locally deployed Ollama or vLLM instance), set the API Endpoint to the local service URL and leave the API Key blank if the service does not require authentication. As long as the service exposes an OpenAI-compatible API, all IDMP AI features work without modification.
