dsh-thinking-effort

When you connect custom or third-party models through llm-pi-ai, Composer sometimes hides the reasoning effort selector because the model entries do not declare reasoningEfforts. Gateway-specific values such as ultra also may not map cleanly to DSH standard levels. dsh-thinking-effort fills that configuration gap, giving manually declared models usable reasoning effort controls.

Its core capabilities include adding default effort options to models that lack declarations, letting you choose per-model levels and exact gateway values from settings, mapping DSH levels to gateway parameters, and setting a default effort for subagents. The subagent default applies only when a request does not already specify an effort, so explicit choices remain respected.

If you mostly use built-in models and their existing reasoning controls are sufficient, this plugin is usually unnecessary. It is a better fit for people managing llm-pi-ai custom models in DSH, wanting fine-grained control over main and subagent reasoning effort, or needing ultra-like gateway values to work inside the standard DSH selection flow.