dsh-sub2api

Developers who subscribe to several AI platforms—OpenAI, Claude, Gemini, Grok—often pipe them through a gateway like Sub2API to get unified OpenAI-compatible endpoints. Until now, those endpoints could not be plugged directly into DeepSeek Harness as model providers. dsh-sub2api closes that gap: it registers your Sub2API gateway as a Harness provider, exposing four platform routes behind a single bare-host base URL. Set a key and the route is live; clear it and the route disappears.

The plugin picks the correct native protocol endpoint automatically based on the key group—OpenAI Responses, Anthropic Messages, or OpenAI Completions—so no path juggling is needed. Streaming, tool calls, reasoning deltas, and token accounting are serialized by the pi-ai adapter, avoiding the tool-call naming misalignments that occur when a gateway force-converts between formats. Built-in model discovery, quota and balance lookups, auto-filled per-model reasoning-effort levels sourced from models.dev, and globally available image analysis and generation tools keep cross-platform workflows seamless inside a single session.

A standout feature is the Auto Vision twin route: every text-only provider route automatically gains a same-name -vision counterpart so any text model can accept image input. The image is transcribed by a configured vision model and the resulting text is fed back to the original model—no session switch or manual model swap required. The plugin is built for developers and research teams who already manage multi-platform subscriptions through Sub2API and want to mix and match models from different vendors, perform image-assisted reasoning, or run parallel tool calls within one Harness workspace.