LM Studio Setup

LM Studio Setup

Use panel and secure companion 1.1.0 or later together. The Grafana LLM app is not required for LM Studio.

1. Load a model and start LM Studio's API server.
2. In Grafana, add an Intelligence Gateway Secure AI data source.
3. Set Provider to LM Studio.
4. Enter a URL reachable from the Grafana server. With Docker Desktop this is commonly http://host.docker.internal:1234/v1; container localhost means the Grafana container itself.
5. For plain HTTP on a trusted local network, explicitly enable Allow insecure HTTP. Prefer HTTPS whenever possible.
6. Set the exact Default model ID and include it in Allowed models.
7. LM Studio commonly needs no credential. Leave API key and bearer token empty unless your server requires one.
8. Select Save & test.
9. Edit the panel, select this secure data source, choose Load models securely, select the approved model, and run Analyze.

Large local models can take several minutes. Keep the panel timeout at or below 600 seconds and align it with the data-source timeout. Reasoning tokens may consume the output ceiling before visible text is produced, so use a sufficiently large administrator and panel cap.

For a suitable local reasoning model, a bounded test example is Maximum output tokens 8192 in both panel and data source, a Requested answer max tokens (soft) target of 500, and 300-second panel/backend deadlines. These are not universal model capabilities. The smaller hard cap applies; selecting provider defaults is not unlimited. There is no separate reasoning-effort control. A model that finishes reasoning without a visible answer needs different investigation from a network timeout.

Images

Load a model that actually supports vision. In the companion, turn on Enable image analysis and add its exact ID to Allowed vision models, as well as Allowed models. Finding an ID in the model list does not verify vision support.

Choose the evidence in the panel:

• Data only: connect Queries → Dashboard → Source panel, leave images unattached, and select Analyze.
• Images only: no query results required; turn Include query data with images off, preview captures or uploads, then select Use images → Analyze images.
• Data and images: connect query results, leave Include query data with images on, preview the images, and explicitly select Analyze images.

Use a synthetic image with known visual facts to verify the response. A model-list request, successful text completion, or mock receipt is not a vision test. See Image Analysis and LM Studio's OpenAI-compatible API documentation.