AI Agent Hub
Back to plugins
🤖

dsh-llm-provider

Model Inference Updated 2026.09.15

Run the following command in DeepSeek Harness:

dsh plugin install imchangchang/dsh-llm-provider

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install imchangchang/dsh-llm-provider in the DeepSeek Harness terminal to install the plugin; the source code is available at https://github.com/imchangchang/dsh-llm-provider .

About this plugin

dsh pins its pi-ai LLM SDK at build time, so every new model added upstream has to wait for a full dsh release. On top of that, the official stack ships a single-DeepSeek model selector and settings page with no multi-provider management, no quota visibility, and no capability labeling. dsh-llm-provider replaces all four of those built-in entries in one plugin, folding model-catalog freshness, provider lifecycle, and quota monitoring into a single unit.

The plugin works along four axes. First, a pi-ai bridge lets the plugin maintain its own copy of the SDK: at startup it pulls the latest upstream version from npm, extracts every sub-path import and named export, builds a probe file, and requires it in an isolated context to verify compatibility; if the check fails it silently falls back to the copy bundled with dsh, with no manual link editing. Second, a fleet of quota adapters calls each provider's free balance endpoint — DeepSeek, Kimi, Moonshot, GLM, MiniMax, OpenRouter, and more — and aggregates the 5-hour and 7-day usage windows onto provider cards and the model-selector trigger. Third, the model selector keeps the official two-level structure (model / reasoning effort) and layers on provider filter chips with quota dots, cross-provider fuzzy search, capability badges, and a model detail card. Fourth, a new Settings tab for model services supports adding providers from presets, running a live connectivity test before writing, storing keys in the dsh credential service, removing and refreshing, and inspecting the current bridge version plus any rejected candidates with reasons.

This plugin is built for users who juggle two or more LLM providers and want to compare remaining quota and switch providers in one place; for developers who want access to newly released pi-ai models without waiting for the next dsh release; and for heavy dsh users doing multi-model orchestration who need a real-time glance at each provider's balance mid-session. All settings are written to the same dsh settings sections and credential store the official stack uses, so there is no data duplication and nothing to clean up on uninstall. The UI strings are currently Chinese only.

Use Cases

  • Manage multiple LLM providers in dsh with real-time quota visibility on each provider card
  • Use newly released pi-ai models in dsh immediately without waiting for a dsh release
  • Switch models and reasoning effort quickly in multi-model workflows using the enhanced selector

Best For

  • Heavy dsh users who juggle two or more LLM providers daily
  • Developers who want access to new pi-ai models as soon as they land upstream
  • AI engineers doing multi-model orchestration who need at-a-glance quota per provider