lanqin996/dsh-reasoning
Per-model reasoning levels and text/image input flags for DSH, edited in a settings card and kept inside the composer's model menu.
Adds a settings card that assigns input modalities and reasoning levels to one or more models, per provider, with bulk editing across a selection. Selection stays in the DSH composer's own model menu, which opens side-by-side so the Model and Reasoning rows remain visible while their options are being browsed.
Install
dsh plugin --profile web add dsh-plugin-reasoning-effortInstall the precompiled npm package into the web profile, then restart the profile so the host and client plugin graph is rebuilt — the package already contains the compiled runtime, so pnpm runs no dependency build script during install and v10 needs no allowBuilds entry. Remove it with dsh plugin --profile web remove dsh-plugin-reasoning-effort. A local checkout is loaded with dsh plugin --profile web add file:. after npm install and npm run build, or through the included cordis.patch.yml.
Compatibility
Changes are written to llm-pi-ai.providers.<provider>.models[].reasoningEfforts and the model-level input field, so DSH and the provider integration validate and send the configured values. An unset input keeps the existing provider or installed-catalog behaviour until you configure it explicitly, and models without configured levels show no reasoning menu at all — the README calls this no fabricated defaults. Standard levels are Off, Minimal, Low, Medium, High, Extra high and Max; custom wire values use id or id=wireValue. It adds no telemetry client, credential flow or background network service, and reads only the active model directory and the llm-pi-ai settings.
Details
- Repo: lanqin996/dsh-reasoning
- Category: Models & Providers
- Stars: 0
- Version: 0.1.9 (npm, LGPL-3.0-only)
Alternatives
karoc/dsh-model-reasoning · HaoyueQin/dsh-better-reasoning-effort · bamboostrip/dsh-reasoning-efforts