DeepCamera is an open-source AI camera platform that runs vision and language models locally on edge devices for video surveillance and security monitoring.
The platform addresses the need for privacy-preserving video analysis by executing inference entirely on local hardware rather than sending footage to cloud services. It combines vision language models like Qwen, DeepSeek, SmolVLM, and LLaVA with object detection, person re-identification, and fall detection capabilities. The architecture centers on a pluggable skill system where each capability is a self-contained module with its own model and parameters, allowing users to compose surveillance workflows from discrete, reusable components. An LLM-powered agent layer adds persistent memory and conversational interfaces, enabling the system to understand scenes, remember observations, and communicate alerts through Telegram, Discord, or Slack.
DeepCamera suits developers building home security systems, NVR deployments, or CCTV monitoring solutions who prioritize local processing and privacy. The platform is designed to run on constrained hardware including Raspberry Pi, Mac Mini, and AI PCs, with automatic hardware detection and model optimization for NVIDIA, AMD, Apple Silicon, Intel, and CPU-only systems. A companion desktop application called SharpAI Aegis removes manual configuration overhead by using an LLM to automatically set up environments, configure skills, and manage the full pipeline without requiring Docker or CLI expertise. The skill catalog currently spans multiple categories with community contributions encouraged through a documented development interface.
Development activity shows active expansion of the skill ecosystem with nineteen skills across ten categories in progress. The project maintains a public roadmap with completed milestones including the skill architecture, a skill store UI, AI-assisted installation, and hardware-aware environment detection. Community engagement occurs through Slack and GitHub issue forums, with the codebase distributed across JavaScript and Python implementations.
