Skip to main content

Heldar

Heldar is a visual event-intelligence operating system for physical spaces. It turns camera streams into structured events, events into workflows, and workflows into operational intelligence. Rather than wrapping an existing DVR/NVR or starting from AI features, Heldar builds its own media kernel first (camera registry, RTSP ingest, recording, playback, live view), then layers perception, a zone engine, and apps on top as consumers. FFmpeg and MediaMTX do the low-level media work; Heldar owns the metadata model, the event engine, and the product logic.

Open-core​

Heldar is open-core:

  • An Apache-2.0 kernel (heldar-kernel) plus generic reference apps (heldar-entry, heldar-movement, heldar-search), a reference composing server, a reference AI worker, and a React dashboard. This is the public heldar repository.
  • Vertical / client products live as separate proprietary crates in a private repository and depend on the open crates. The kernel never references them.

Apps plug into the kernel only through a small set of public seams, so the kernel has no dependency on any app. A deployment is composed from the kernel plus whichever apps a client needs (single-tenant per deployment). See Open-core for the boundary and Architecture for the seams.

Architecture at a glance​

Heldar open-core architecture

The kernel is the only component that talks to cameras. The 24/7 recorder keeps the compressed stream decode-free; a budgeted sampler is the only thing that decodes, writing one current frame per camera. AI workers are pure HTTP clients: they pull sampled frames and post detections back. Apps interpret those detections into domain events.

Where to go next​

  • Quickstart - build, run, add a camera, and run the AI worker.
  • Deploy - one binary, one URL (Docker one-liner, native binary, or a flashed appliance).
  • Remote Access - view a site from anywhere in a browser, over WebRTC, even behind CGNAT.
  • Using the Dashboard - a tour of the web UI: live view, playback, zones, incidents, and the System page.
  • Architecture - the kernel and its four public seams.
  • Open-core - what is open versus proprietary.
  • Build a module - build your own app against the open kernel.
  • Build an AI worker - the perception worker SDK contract.
  • Operate - the in-repo operator and integrator guides.

Source lives at github.com/Straits-AI/heldar.