Skip to content
Emre Yavuz edited this page Aug 2, 2026 · 2 revisions

Overseer

Overseer is an offline, single-camera AI surveillance command center. It turns one ordinary video feed into a real-time understanding of a scene: who and what is present, where they are in 3D, how they move, what is unusual, and who they are across time. A natural-language AI Operator lets you run the whole system by voice or text.

Everything runs on the local machine. No cloud calls, no footage leaving the host. Models are vendored and cached locally, and the platform runs air-gapped.

Product UI and documentation are in English. The stack is a Python (FastAPI) backend plus a Svelte 5 / Electron frontend, packaged as desktop installers for Windows, macOS and Linux.

What it does at a glance

Area Highlights
Perception YOLO11 detection, ByteTrack tracking, segmentation masks, monocular depth (Depth Anything V2), fine-grained attributes (vehicle body type, brand, colour; person stature, clothing colour, behaviour)
Spatial & 3D Depth-lifted point cloud and mesh, background completion, a fly-through 3D viewport, foresight (path prediction), tactical god-view radar, predictive ghosts
AI Operator Voice + text control of the entire app, 90+ chained actions with data passing, offline speech-to-text (Turkish + English), spoken replies, and vision Q&A about the live frame
Identity Cross-camera re-identification, long-term dossiers, gait descriptors, a relationship graph, ANPR licence plates, watchlist / BOLO
Analytics Zones, alert rules, correlated threat scoring, smart alert + zone suggestions, per-camera DNA, live statistics
Forensic Attribute search, search-by-image, plate search, and a live "enhance" loupe

Start here

  • Getting Started - install the desktop app or run from source, and first-run notes.
  • Architecture - how the backend, frontend and data flow fit together.
  • Models - what models are used, how they download, and manual model steps.
  • Configuration - every knob in config/default.yaml and the environment overrides.

Feature guides

Reference

Links

Clone this wiki locally