dsh-llm-commandcode

DeepSeek Harness ships with a fixed set of built-in inference channels, but dsh-llm-commandcode adds a dedicated Command Code Provider API route so you can plug chat, tool use, and streaming reasoning into Command Code's official endpoints without swapping the host.

Three capabilities stand out. First, full chat protocol support through DSH PiAiAdapter: claude-family models speak Anthropic Messages while every other model speaks OpenAI Chat Completions, with SSE, tools, attachments, and reasoning all flowing over the same pipe. Second, model discovery and catalog management: one click pulls the Go / Pro / Provider+ layered model list, with drag-to-reorder, context-window overrides, and official effort selectors per row. Third, credential and quota handling: the API key lives exclusively in the Host credentials service, and account quota (plan, credits, 5-hour and weekly windows) is shown on a best-effort basis that never blocks a chat request. An optional zero-data-retention toggle appends x-cmd-zdr: 1 at the request level.

It is built for developers who already hold a Command Code API key and want it as a day-to-day inference backend inside DeepSeek Harness, as well as teams that need multiple providers side by side and want to swap models or protocols per task. In headless mode the Host route works with no Web UI required; in a Web setup, pairing the plugin with dsh-llm-providers-ui unlocks the full Settings card experience.