DeepCamera is an AI-powered camera skills platform that runs computer vision and vision language model inference locally for privacy-sensitive surveillance. It combines object detection, face recognition, person re-identification, and VLM-based scene analysis using models like Qwen, DeepSeek, SmolVLM, and LLaVA. The system operates as an LLM-powered agent that monitors cameras, understands scenes, and sends alerts through Telegram, Discord, or Slack. All inference runs locally using llama.cpp on consumer hardware like Mac Mini and AI PCs.