dsh-effort-tweak

When using DeepSeek Harness, model capabilities often need to be edited in the llm-pi-ai configuration file, but the official Web UI may not yet expose the necessary controls. This can be inconvenient for users who prefer to adjust model behavior through a graphical interface. The dsh-effort-tweak plugin addresses this issue by adding a dedicated settings page for reasoning effort levels.

At its core, the plugin introduces a 'Settings → Reasoning effort' page within the DSH Web UI. It lists models explicitly configured in custom providers and allows users to choose reasoning levels—such as off, minimal, low, medium, high, xhigh, or max—for each model individually. This enables fine-grained control over model reasoning depth and resource usage, with settings saved automatically via DSH's settings service, eliminating the need for manual configuration edits.

dsh-effort-tweak is ideal for users who need to flexibly adjust model reasoning behavior, particularly developers, researchers, or advanced users. It benefits those who use custom providers to integrate specific models and want to quickly test different reasoning levels through an intuitive interface, bypassing the complexity of direct configuration file changes. The plugin streamlines this process, enhancing usability and efficiency.