SideSight: A Visual Sidecar for Text Coding Agents

SideSight is a CLI-first visual sidecar feature of DeepSeek Harness that allows agents to analyze visual content, such as screenshots and charts, using multimodal models. It supports file-system skill integration, an optional MCP server, and native offline OCR on macOS. It can be installed either as a DSH plugin or a file-system skill, and its configuration requires specifying an OpenAI-compatible vision model. When using it, image paths must be explicitly specified in the instructions, and API keys are stored only in the configuration file.