본문으로 건너뛰기
버전: dev

LM Studio monitoring

This guide describes the existing Agentless monitor template for LM Studio model metadata. It does not configure LM Studio as an AI Gateway provider, instrument inference requests or establish model quality/performance.

Configure the monitor​

  1. Start the LM Studio HTTP service and verify that it is reachable from the selected HertzBeat collector. A collector running in another host or container has its own localhost.
  2. Select LM Studio in the monitor creation catalog. Set Host, Port (template default 1234) and HTTPS to match that service.
  3. If the endpoint requires authentication, set the monitor's API Token. The template uses Bearer authentication. Keep credentials in protected configuration and out of screenshots and diagnostic exports.
  4. Run the form's detection, inspect the returned model fields, then save. Recheck current samples after at least one configured collection interval.

The source template hertzbeat-manager/src/main/resources/define/app-lmstudio.yml sends GET /api/v1/models and reads models. The matching upstream contract is documented in LM Studio's model-list API. An older or different endpoint is not automatically compatible with this template.

What the rows mean​

The template exposes model key, display name, type, publisher, architecture, quantization, model size, parameter description, maximum context length and format. Model size is converted from bytes to MB by the monitor configuration. Optional or absent model fields must not be read as measured zeroes. This is an inventory response, not per-request latency, token throughput or GPU utilization.

Diagnose an unsuccessful collection​

Check host reachability and the exact port/protocol before changing credentials. A 401/403 response requires an authentication/permission correction; it is not an empty model list. For a parsing failure, compare the endpoint and response field names with the source template. An empty models array can be valid when there are no available models; it does not prove inference availability.

This alpha acceptance validated the Agentless MySQL and OTel Java paths, not a real LM Studio installation. The template and upstream API were reviewed; test your LM Studio version and retain the actual result before treating it as a verified source. For provider setup, use the separate AI Gateway guide.