dsh-bailian-models

DSH's built-in pi-ai adapter does not recognize the dashscope.aliyuncs.com endpoint. Requests silently fall back to the plain OpenAI dialect, enable_thinking is never sent, and model entries carry no context-window or reasoning metadata—so thinking depth is locked and context size is a guess. dsh-bailian-models exists to close that gap: it injects a ready-to-use Alibaba Bailian (DashScope) inference route with a built-in catalog of 36 major text models, each annotated with context window, max output tokens, and a per-family reasoning-effort schema (effort levels, token budgets, simple toggle, or always-thinking). Once installed, Bailian models appear in the DSH model picker with working reasoning controls—no hand-written settings.yaml required.

Beyond the preset route, the bundle mounts an auto-adapter that watches your existing provider routes. If any route's baseURL resolves to a Bailian host (dashscope.aliyuncs.com, dashscope-intl, dashscope-us, or *.maas.aliyuncs.com), the adapter fills in the missing dialect compat flags, reasoning-effort declarations, and per-model capacity fields. The rule is strict: only gaps are filled, values you wrote yourself are never overwritten, and unknown model ids are left completely untouched. Every write is validated against pi-ai's own schema before landing; a rejected write leaves your configuration exactly as it was and logs a warning.

If you call Qwen, GLM, DeepSeek-V4, Kimi, MiniMax, or other Bailian-hosted models inside DSH and want to dial thinking depth the way a native adapter would—scale reasoning effort, check context budgets, and safely turn thinking off to control cost—this plugin is built for you. Zero runtime dependencies, MIT licensed, and uninstalling it leaves no harmful residue in your settings.