dsh-smart-scenario-router

Running GLM, DeepSeek, Qwen and GPT side by side in DeepSeek Harness means juggling model choices with every prompt: you want GLM for architecture planning, DeepSeek Pro for deep code review, and Flash for firehose batch extraction. Manually switching between them breaks flow and is easy to get wrong.

dsh-smart-scenario-router removes that burden. It reads each user message, classifies it into one of ten scenarios covering architecture design, code review, multimodal image-text input, long-context analysis and more, then picks the first available model along a priority chain. If the upstream call fails, it slides automatically to the next candidate until the request succeeds. Messages with low confidence are re-judged by a designated referee model (Flash by default), and the full model pool, per-scenario chain and live routing state are all visible and editable right in the dsh settings page. The plugin respects manual model choices and never touches API keys.

If you have at least two domestic models plus a GPT fallback configured in dsh and your daily work oscillates between coding, planning and batch extraction, this plugin lets you stop thinking about which model to pick. A restart after install is all it takes.