Multi-camera intelligence

Efficient enough to run anywhere

Inquire for Production Deployment →
01 // Compression Engine

INT8 Quantization

Detection models are quantized with Kenosis and verified against COCO ground truth. The verification measures detection accuracy itself, not a proxy metric.

02 // Throughput

Lock-Free Pipeline

Every camera runs in its own isolated thread. At real-time tiers the pipeline drops zero frames, and telemetry counter-verifies it.

03 // Reliability

Memory-Safe Core

The pipeline is native Rust and memory-safe by construction, with no Python layer in the serving path.

04 // Routing

Hardware Acceleration

The same build needs no GPU, and it picks up CUDA and DirectML acceleration when the host has one.

05 // Observability

Production Telemetry

Prometheus and REST monitoring are built in, with per-stage latency histograms and overload alerting. Every performance claim on this page is reproducible from the daemon's own counters.

Measured — multi-stream headless daemon, 720p sources, 2026-07-21

30 fps
per camera, two streams, full rate
97 / sec
detections across four streams
2.5 cores
total CPU at the two-stream rate
40.6 mAP
six streams at the accuracy tier

Deterministic 720p30 streams · PP-YOLOE+ S INT8 · Intel i5-13420H · 45 s windows · every configured stream verified at full ingest rate

Measured — multi-camera scaling, deterministic clip matrix 2026-07-21

97 fps
peak aggregate inference, four cameras at 320
90 fps
aggregate across six cameras at 320
24.2 vs 13.1
INT8 vs FP32 aggregate fps · six cameras at 640 on OpenVINO, ~7 cores each
40.6 mAP
COCO at 640 · the same score on ONNX Runtime and OpenVINO

Four to six cameras · 720p30 fixture clips over RTSP · PP-YOLOE+ S · 2 threads per camera · Intel i5-13420H · 45 s steady-state windows · FP32 shown on its faster runtime