AI Agent Hub
Back to plugins
🖥️

dsh-plugin-lmstudio

Client Updated 2026.08.16

Run the following command in DeepSeek Harness:

dsh plugin install starlightzfy/dsh-plugin-lmstudio

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install starlightzfy/dsh-plugin-lmstudio in your terminal; the plugin repository is at https://github.com/starlightzfy/dsh-plugin-lmstudio. After installing, restart dsh web and pick your LM Studio local model in the model picker to get started.

About this plugin

The DSH model picker defaults to cloud APIs, yet local inference should never burn through your quota. This plugin seeds DSH's built-in llm-pi-ai adapter with LM Studio's OpenAI-compatible endpoint (127.0.0.1:1234/v1 by default). Install, refresh the page, and your local model shows up in the picker. All traffic stays on localhost; your DeepSeek balance is untouched.

It writes a default provider route as a base layer (baseURL, wire protocol, example model). Anything you add under the llm-pi-ai section of settings.yaml merges over it per provider, so your own models and endpoints always win. No API key is needed for a local server; if you enable key auth in LM Studio, set apiKeyEnv in your settings layer. settings.yaml hot-reloads, so no dsh web restart is required.

Ideal for developers who have a model loaded in LM Studio and want it as a first-class chat model in DSH, for anyone who wants to trial different local models without switching toolchains, and for users who prefer plain conversation without agent tools: create an empty agent preset and pick it on the new-session screen.

Use Cases

  • Use an LM Studio local model directly in the DSH chat UI with zero DeepSeek API consumption.
  • Switch between multiple local models for quick A/B testing; settings.yaml hot-reloads without a restart.
  • Keep all inference on localhost for privacy-sensitive or air-gapped workflows.

Best For

  • Developers who have models loaded in LM Studio and want a seamless DSH chat experience.
  • Users who want to trial local models at zero cost without changing their toolchain.
  • Users who prefer lightweight, tool-free plain conversation over agent workflows.