A platform you own, end to end
Own the kernel
Heldar owns its own media kernel: camera registry, RTSP ingest, recording, timeline, and live view. You own the metadata model and the event engine, not a slot in someone else’s VMS.
AI perception built in
Detection, tracking, and zones out of the box, with ANPR access control, cross-camera movement, and semantic search as first-class consumers of the event stream.
Open-core and extensible
An Apache-2.0 kernel you can read, run, and extend. Write your own module against the DetectionConsumer seam and compose it into the deployment you need.
Your hardware, your data
Runs on-prem on your own boxes. Browser-based WebRTC remote access keeps operators connected with the video end-to-end encrypted — only signaling and relay are hosted, never your footage.
One platform, from packets to answers
The kernel records and indexes 24/7, perception layers on as a consumer, and apps read the same event stream. Every capability below ships in the same codebase.
Recording and DVR
FIFO retention, evidence-lock, HLS playback, plus backup and archive of the footage that matters.
Detection and zones
Tracked detections feed a polygon zone engine that raises enter, exit, and dwell events with evidence frames.
Access control and ANPR
Per-frame plate reads consolidated by temporal voting into entry and exit events, resolved against a registry.
Movement and ReID
Cross-camera correlation over an operator-defined topology graph, where every link is a candidate a human confirms.
Semantic search
Natural-language questions become deterministic query plans over your stored event facts. Works fully offline.
Alerting and webhooks
Warning and critical events delivered to your endpoint at-least-once, decoupled from the recording path.
Camera configuration
Onboard and manage cameras over ONVIF and ISAPI directly from the dashboard, with credentials kept server-side.
One-binary deploy
A single Rust binary serves the API and the bundled dashboard. One process, one URL, your hardware.
Build on the open kernel
Heldar is a platform, not a black box. The kernel is Apache-2.0, the seams are documented, and the reference apps show you the patterns. Write a perception worker or a new app against the DetectionConsumer seam and ship it on your own terms.