Deepseek Harness Vision Plugin
sjscy05/deepseek-harness-vision-plugin · model-adapters
is stronger) OpenAI-compatible OpenAI-compatible OpenAI-compatible Anthropic Messages API ( base64 blocks) Google Gemini ( parts) zhipu / qwen / doubao / openai share the OpenAI-compatible protoc.
★ 2GitHub stars
0Forks
2026-08-25Last updated
TypeScriptLanguage
MITLicense
Key features
- image (required)a local file path (absolute, or relative to the harness working directory), an http(s) URL (downloaded with the same timeout budget), or a data:image/...;base64,...
- Supported formatsPNG, JPEG, WebP, GIF.
- questionthe specific question about the image — phrased from the user's request or the current task's need (e.g.
- "What text is visible?", "Describe the chart's trend"), not a generic "describe this"
Requirements
- The vision sub-model is selected by config.provider — six mainstream vendors are built in, switching is one line
- zhipu / qwen / doubao / openai share the OpenAI-compatible protocol
Install command
dsh plugin --profile web add github:sjscy05/deepseek-harness-vision-plugin