dsh-performance-slider

Switching between models and reasoning-effort tiers in DeepSeek Harness used to mean opening a settings panel, scrolling past unrelated options, and double-checking what you actually changed. dsh-performance-slider collapses that into a single gesture: a Codex-desktop-style 6-level slider that sits directly above the input box. Drag left for a lightweight flash model at minimal reasoning, drag right for the pro tier at maximum effort. Every notch locks in a model plus a reasoning budget in one move, eliminating the cascade of toggles.

While you drag, the background scrubs continuously through 31 locally hosted character cutout frames; release and it snaps to the nearest anchor (frames 0, 6, 12, 18, 24, or 30). A low-opacity theme tint keeps the character visible without hurting text or code readability. Your selection persists in both browser localStorage and a port-independent host-scoped cookie, so even when dsh web launches on a random port each time, the background and model level restore automatically on your next session.

For developers who want to reshape the tiers, the LEVELS table in lib/client.frames.js lets you bind any provider model id per level, with a secondary modelClass (flash / pro) loose match so date-suffixed custom models still resolve correctly. Reasoning effort follows a preferred-then-fallback chain and ultimately falls back to the model-directory default. You can also open demo/index.html in a plain browser to experience the full drag-and-snap interaction without installing DSH at all. This plugin is for dsh web and dsh-desktop users who switch models and reasoning intensity often and prefer one tactile drag over a stack of setting changes, as well as developers who want a little visual personality in the Harness background.