dsh-vision-better 1.0.1

Multi-provider vision backends for DeepSeek Harness.

New in 1.0.1

• Provider/protocol selection: OpenAI-compatible (Zhipu, reverse proxies, relays, self-hosted), Anthropic, and Google Gemini native protocols.
• Model catalogue fetching through the draft values in the settings page (never persisted behind your back).
• Switching provider fills in that protocol's default endpoint and model.
• Config-file diagnostics: an unreadable or invalid config is explained instead of silently showing defaults.
• Launch-environment API keys are read-only and explained, instead of a raw provider refusal.
• The OpenAI-compatible 400 variant walk (bare base64, omitted thinking) applies only to that protocol.

Install

dsh plugin --profile web add github:ethanweave/dsh-vision-better

Important: image admission for a text-only model needs the small core extension in patches/ - see the README (Core extension).

Verified against DeepSeek Harness 0.2.0-rc.2.

dsh-vision-better 1.0.0

Vision relay for DeepSeek Harness: when the active model cannot accept images, a configurable GLM vision endpoint reads them in the background and the original model keeps answering in the same session.

Install

dsh plugin --profile web add github:ethanweave/dsh-vision-better

Important: image admission for a text-only model needs the small core extension in patches/ - see the README (Core extension). Without it the settings page works, but image turns are refused by stock DSH.

Verified against DeepSeek Harness 0.2.0-rc.2.