| name | vision-analyze |
| description | Capture and analyze a scene from the connected camera. |
Vision Analyze
Look through the connected camera and analyze the scene using vision AI.
Inputs
- prompt: What to look for or analyze (optional, default: general scene description)
- mode: Analysis depth — "quick" for detection only, "full" for VLM scene analysis (optional, default: "quick")
Execution
- Capture a snapshot and run detection:
look(prompt="<PROMPT>")
- If people are detected and identification is needed:
who_is_here()
- Report findings with detected objects, recognized faces, and scene description.
Rules
- Always describe what is visible before interpreting
- If unknown persons are detected, mention it but don't escalate unless asked
- Include confidence levels for object and face detections
- Respect privacy — don't store or share images without explicit instruction