@kingsunb/dsh-model-plus
DSH Web plugin: on the Settings page, under "模型 Plus" (Models Plus), edit each model's reasoning-effort tiers, vision capability, context window, and default output limit per provider, entirely in your local configuration. One-click sync fills in missing fields from models.dev by model id.
Features
- Per-provider, per-model configuration
- Reasoning-effort tiers (off / minimal / low / medium / high / xhigh / max) - Vision capability (input: text / image) - Context window contextWindow (tokens) - Default output limit maxTokens
- Quick presets: apply common combinations in one click (reasoning off / basic 3-tier / all-open / vision + all-open)
- models.dev sync: fill in reasoning tiers, context, maxTokens, vision by model id (only fills missing by default; can overwrite)
- Add providers: URL + API key import, fetch model lists, auto-fetch on creation
- Model testing: single or batch tests, customizable default prompt, SVG preview
- Multi-provider switching: manage multiple providers on one settings page
Installation
dsh plugin --profile web add @kingsunb/dsh-model-plusRestart dsh web, then open Settings → 模型 Plus.
Catalog sources
The sync settings page provides three choices:
- Official models.dev:
https://models.dev/api.json - China GitHub acceleration:
https://gh-proxy.org/https://github.com/kingsunb/dsh-model-plus/raw/refs/heads/main/api.json - Custom address: any approved HTTPS catalog URL
The China source serves the api.json snapshot committed at this repository root through gh-proxy.org. The official models.dev source remains the default.
Architecture
- Host half (
lib/index.js): registers same-origin HTTP endpoints/api/plus/*
- bootstrap / list-models / save-model / apply-preset - add-provider / discover-models (models.dev enrichment when fetching models) - enrich-models (one-click sync / sync settings page: models.dev preview and apply) - test-model (model availability and creative output testing)
- Client half (
lib/client.js): registers the "模型 Plus" settings section
License
MIT