dsh-codebuddy-models

If you already rely on a Tencent CodeBuddy or WorkBuddy desktop subscription and wish to call those models directly from DeepSeek Harness, the previous recipe meant standing up a Python conversion proxy, hand-rolling auth scripts, and juggling two disjoint runtimes. dsh-codebuddy-models folds all of that into a single self-contained TypeScript plugin: credential reading, automatic token refresh, direct backend calls, and SSE stream parsing are built in, with zero external converters or extra runtime dependencies.

The plugin reuses the desktop client's existing login state, stores no passwords, and performs no independent authorization. The upstream endpoint speaks the standard OpenAI chat-completions protocol, so function calling and tool_calls work out of the box. At build time esbuild inlines every runtime dependency into one artefact, allowing the plugin to load cleanly even inside DSH Desktop's preset-plugins directory where node_modules is absent. A dedicated CodeBuddy Models section appears in the DSH web settings page, presenting a live enterprise model catalogue (name, input/output capacity, description), exposing tunable request parameters such as endpoint, context window, and idle timeout, and offering a collapsible fallback model list that activates only when auto-fetch is unavailable, all applying live on save.

It is aimed at developers whose daily workflow depends on CodeBuddy or WorkBuddy while they simultaneously run DeepSeek Harness as their primary agent layer, wanting both model families available in a single picker within the same pipeline. No extra accounts to register, no proxy servers to babysit, no Python environment to maintain. Install the plugin, refresh the GUI, and the CodeBuddy models are right there.