DSH Vision Toolkit adds native vision capabilities to DeepSeek agents
It gives text‑only DeepSeek Harness agents visual perception by wrapping the agent‑vision‑toolkit as a native profile bundle. The toolkit offers intent‑aware image Q&A, long‑screenshot OCR, UI reconstruction, pixel diff, and artifact management through structured schemas. It targets developers building AI agents that need vision, providing seamless runtime preparation and web UI instead of fragmented CLI scripts. Compared to ad‑hoc glue code, it delivers tighter integration, richer tooling, and easier agent interaction with visual data.
View on GitHub →Anionex/dsh-vision-toolkit