dsh-model-aliases

Switching models in DeepSeek Harness usually means walking through provider, model, and reasoning effort in sequence; with several providers and multiple reasoning tiers configured, every switch feels like digging through a nested menu. dsh-model-aliases collapses that entire chain into a single memorable name right beside the input box, so one click does what three dropdowns used to.

An alias is more than a shortcut. The settings panel lets you add, reorder, and remove entries centrally, each one pinning a provider, a model, and an optional reasoning level while automatically surfacing only what the current model directory actually offers. Configurations whose target has temporarily disappeared from the directory are kept rather than deleted, shown as disabled with a reason. The selector hides itself when a session is locked or when the conversation is addressed to a sub-agent, preventing accidental switches. Choosing provider default for reasoning effort writes no fabricated value, preserving the adapter's native behavior. Everything persists in the DSH model-aliases settings namespace and survives restarts; saving an empty list even restores the preset aliases automatically.

Built for heavy users who juggle multiple providers and reasoning tiers in Harness, especially those maintaining distinct workflows for everyday chat, long-form reasoning, and code assistance and would rather manage model configs by name than by a three-level path.