vision-tools

The codex-deepseek-vision project installs the following CLIs on this machine. They share the same vision config as the proxy (VISION_API_KEY / VISION_BASE_URL / VISION_MODEL / LANG) — no extra credentials needed:

• glance: image description, Q&A, and OCR
• ground: locate targets in an image with natural language and get bounding boxes in original pixel coordinates

glance

glance <image>                   # describe the image in detail
glance <image> -q "<question>"   # ask a question about the image
glance <image> --ocr             # verbatim OCR of all visible text

Output language follows the LANG setting in the vision config (zh/en).

ground