dsh-llm-opencode-go

DeepSeek Harness already supports multiple model providers out of the box, but OpenCode Go models such as Muse Spark and GLM-5.3-Flash had no first-class integration. dsh-llm-opencode-go closes that gap by bringing the full OpenCode Go model catalog into Harness unified routing layer, so you can call these models without leaving your existing workflow.

The plugin key strength is per-model protocol selection: each model is matched to the correct endpoint (Completions, Responses, or Anthropic Messages) from the official Go catalog table automatically, with no manual per-model configuration. Model discovery fetches the live model list and enriches each entry with context window, vision, and thinking metadata, while subscription usage reads the 5-hour, weekly, and monthly consumption meters from native Go endpoints. Both capabilities stay outside the chat protocol. The model catalog is fully interactive: drag to reorder, expand a row to inspect capability flags, and remove entries you do not need. API keys are write-only and never echoed to the browser.

Best suited for teams or individual developers who already manage multi-provider models in DeepSeek Harness and want OpenCode Go on the same routing and usage dashboard. In Web mode, pair it with dsh-llm-providers-ui for the complete LLM Providers settings page; in Headless mode the Host route works standalone without the UI owner package.