Design-fidelity QA for DeepSeek Harness: lend any text-only model an eye, then judge whether the implementation matches the mock. Ships the benchmark behind that judgement — four fixtures, 23 injected defects, and every raw model transcript. Retires itself when DeepSeek ships vision.
DeepSeek Harness all-in-one: no model switching — regular DeepSeek auto-routes to vision & image gen. Multi-backend: Gemini + any OpenAI-compatible (GPT-4o, Qwen-VL, GLM-4V, gpt-image, DALL-E, Flux, OpenRouter). gemini_vision/gemini_generate_image/gemini_optimize_image with vision self-check. Better than modlens.
Give your DSH agent eyes via any OpenAI-compatible vision model - 11 provider presets (Doubao/Qwen-VL/GLM-V/OpenAI/Gemini/Ollama...), capability checkboxes that inject live prompt guidance, and analysis of images the user drops into the chat; the agent writes its own observation prompt, and base64 never enters its context.
