dsh-cloud-model-providers

If you are already running DeepSeek Harness and want Ant Digital MaaS or NVIDIA NIM models to sit directly inside the DSH runtime, this plugin exists for exactly that. It does one thing: it injects three cloud model routes into an existing base-backed profile via a single declarative cordis.patch.yml. No extra HTTP proxy, no build scripts at install time, and no API keys stored anywhere in the repository.

The three routes cover Ant Digital MaaS OpenAI Responses over SSE, Ant Digital MaaS Chat Completions over SSE, and the NVIDIA NIM catalog-backed Chat Completions over SSE endpoint. Every piece of plumbing that used to require a separate gateway-protocol translation, SSE chunking, tool-call relay, usage accounting, request cancellation, and bearer-credential injection-is handled by the llm-pi-ai adapter that ships with DSH itself. You set MAAS_API_KEY or NVIDIA_API_KEY through environment variables or the Web settings page, pick a model, and the maintained DSH path carries the rest.

This is aimed at developers who live in the DSH ecosystem, genuinely need to call Ant Digital or NVIDIA hosted inference in their day-to-day workflow, but do not want to stand up and babysit an extra gateway or sidecar process. One patch layer in, restart the profile, and the routes show up in the Models page. When you are done, remove the same layer and its dependency together and nothing lingers behind in your profile.