HaoyueQin/dsh-better-reasoning-effort
dsh-better-reasoning-effort brings reasoning-effort and input-modality editing for third-party models back into the DeepSeek Harness UI: an Options block inside each model row on the official Models page, a one-click Auto-adapt knowledge base of 65 curated entries across 15 vendors, and a reasoning-effort slider inside the official composer model menu.
dsh-better-reasoning-effort brings reasoning-effort and input-modality editing for third-party models back into DeepSeek Harness. The llm-pi-ai adapter natively supports per-model reasoningEfforts and input declarations, but the official Models page editor keeps both fields out of reach, so third-party models get no thinking-level picker in the composer and hand-declared models are treated as text-only. The plugin injects an Options block under each model row's disclosure (reasoning effort, input modalities, endpoint compatibility) that commits with the card's own Save; a one-click Auto-adapt fills recommended levels, wire spellings and modalities from a built-in knowledge base of 65 curated entries across 15 vendors plus a same-origin /models probe; and the official model menu's body is replaced by an upstream-style reasoning-effort slider with per-session memory. It also adds a per-model default effort stored in the settings document, a provider-card request-header editor with a fetch-layer user-agent override, and three intents (all levels off = unset, only off = disable reasoning, levels armed = write the declaration), all keyed off the official page's DOM so an official upgrade merely pauses injection.
Install
dsh plugin --profile web add dsh-better-reasoning-effortCopied verbatim from the README's npm install block; the README also documents a version-pinned form (dsh-better-reasoning-effort@0.5.2), a release tarball URL, a github: source form and a link: local-checkout form.
Compatibility
Requires DeepSeek Harness 0.1.5-alpha.1 or later (the 0.1.2-rc / 0.1.3-alpha lines are no longer supported; use plugin 0.3.7 there); compiled and gated against 0.2.0-rc.2 and the 0.2.1-alpha line. The composer slider is adapted from HanaAyane's dsh-reasoning-effort - remove that plugin to avoid two effort controls on the same seat. English README with a Chinese translation.
Details
- Repo: HaoyueQin/dsh-better-reasoning-effort
- Category: Models & Providers
- Stars: 50
- Version: GitHub repo HaoyueQin/dsh-better-reasoning-effort (50★, MIT); npm package dsh-better-reasoning-effort at 0.5.3 (registry check 2026-10-13; the registry repository field resolves back to this repo).
Alternatives
aa2246740/dsh-antigravity-oauth · AmmarByFar/dsh-trustedrouter · artorias-zj/dsh-opencode-free-model