dsh-provider-pro

Hand-adding custom providers through llm-pi-ai in DeepSeek Harness silently disables the official channels thinking-level picker and request-header customization, and there is no quick way to verify what each model actually supports. dsh-provider-pro closes that gap: it auto-populates a five-level reasoning dictionary for every hand-declared custom model so the in-chat dropdown works exactly like the built-in channels, with zero per-model setup.

Four capabilities work together, each solving a concrete pain point. A per-provider User-Agent card lets you override the built-in attribution header to bypass rate-limiting or non-official-app blocks. A single checkbox on each model card declares image-input support so DSH immediately allows image attachments. The built-in one-click full probe measures context window, message-role compatibility, real reasoning-level acceptance, and image-input latency in a single pass, auto-removing any level the upstream rejects so the picker only ever offers what the provider truly accepts.

This plugin is built for heavy DSH users juggling multiple LLM endpoints like GLM relays, self-hosted OpenAI-compatible gateways, or third-party services with UA-based throttling who want reasoning switching, image attachment, and model liveness checks to be as seamless as the official channels, without hand-tuning every single model card.