Argus

i've always been super fascinated by surveillance systems and tactical command centers—not in a creepy way, but just how complex data gets correlated in real-time. argus is my personal, miniature version of palantir gotham that runs completely on my own local hardware. normally, security cameras just dump raw, compressed mp4 streams onto a hard drive. nobody actively watches them until something goes wrong, and even then, you have to scroll through hours of footage manually. argus actively watches and thinks about the video feed. it uses a multi-threaded python pipeline that ingests RTSP streams from local cameras using FFmpeg. we feed these streams into local, optimized YOLO models for object detection and run simple tracking algorithms (like SORT) to follow entities across frames. the magic happens in the correlation engine: if camera A detects a person walking east, and camera B detects the same visual signature ten seconds later moving towards the main door, argus links those logs as a single event. it stores everything in a local database and renders it onto a dark, tactical dashboard. the best part is that it is 100% local. no cloud APIs, no subscriptions, and zero data leaving my room. palantir built theirs with hundreds of engineers and government funding; i built this because i wanted to see if i could do it alone.