dsh-model-sync

When you configure multiple LLM providers in DeepSeek Harness, their model lists are rarely static: models get deprecated, new ones appear, and the locally stored list quietly drifts out of date. Manually checking each provider's /v1/models endpoint whenever something changes is tedious and easy to forget. dsh-model-sync automates exactly that.

The workflow is straightforward. Open Settings → Model Sync and click Sync All Models for an instant refresh of every configured provider, or flip on Auto-sync on conversation open so the list refreshes each time a conversation page loads (10-second debounce built in). Sync uses replace semantics: whatever the endpoint advertises becomes the final list — removed models disappear, added models appear, and all other provider fields are preserved. Failure isolation keeps things safe: if one provider's request fails it does not affect the others, and the failed provider simply keeps its previous configuration, with a clear ❌ reason shown in the panel.

If your setup spans several LLM providers and you would rather not manually verify model changes each time, this plugin is a low-friction addition. It relies solely on the official wire APIs (llm.discoverModels + settings.mutate), introduces no custom Host RPC, and leaves a minimal footprint with no side effects.