AI Agent Hub
Back to plugins
🤖

cedardsh-model-probe

Model Inference Updated 2026.08.31

Run the following command in DeepSeek Harness:

dsh plugin install maiziman/cedardsh-model-probe

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

Run dsh plugin install maiziman/cedardsh-model-probe inside DeepSeek Harness to install; the plugin source is available at https://github.com/maiziman/cedardsh-model-probe , and once installed, capability fields for your custom compatible endpoints are auto-detected and populated in settings.

About this plugin

When you connect a custom OpenAI-compatible endpoint to DeepSeek Harness, figuring out which reasoning levels it actually supports, or whether it accepts image input at all, is often left to guesswork. CedarDSH Model Probe eliminates that guesswork entirely: after a one-step install, you simply add or save a compatible provider and the capability fields quietly populate themselves in the background.

The plugin first reads capability metadata supplied by OpenRouter, Ollama, and other compatible endpoints. When that metadata is missing, it sends a single lightweight probe to local or private-network endpoints only, using a fixed arithmetic prompt plus two 32x32 color test images, to verify low, high, and max reasoning levels as well as image-input support. Any capability value explicitly set by the user or provider always takes precedence, and public endpoints stay metadata-only unless the user changes the probe policy. Because the plugin never touches the official DSH adapter, future Harness upgrades will not overwrite anything.

Who it is for: if you regularly attach custom models to DeepSeek Harness through Ollama, vLLM, OpenRouter, or similar endpoints and would rather not dig through documentation to guess capability boundaries, this plugin is a no-fridge, no-config-file, no-window background helper that simply makes model capabilities visible and correct.

Use Cases

  • Attaching a custom Ollama or vLLM model to DeepSeek Harness and letting capability fields populate automatically
  • Connecting an OpenRouter or other public endpoint and getting supported capabilities without reading documentation locally
  • Spin up a private model server and confirm low, high, and max reasoning levels with a single lightweight probe request

Best For

  • DeepSeek Harness users who connect custom endpoints via Ollama, vLLM, or OpenRouter
  • Local LLM operators who prefer automatic capability alignment over manual documentation lookup
  • Advanced users who need distinct probe policies for public versus private endpoints and want explicit settings to always take precedence