Bring your own models
Prepare your model for the platform. Configure inputs, labels and thresholds, with video and events built in.
Explore model integration →Your modelBuilt for your platformActual analysis ↗
CosmoEdge / Open-source edge vision AI
Turn video into
actionable events.
Connect video, models and scenario rules to bring intelligence into your workflow.
Vision algorithms. Already included.
- Safety helmets
- Fire & smoke
- Falls
- People & vehicles
02Logic & rules
Detect peopleClassify helmetsEvaluate rules03Event capture
No safety helmetView event Video inputConnect your on-site streams
Edge analysisRun vision models and scenario rules
Business eventsDeliver structured events to your systems
01 / A complete scenario
Helmet detection, from setup to alerts.
Set a detection region at a construction entrance, configure the helmet rules, then view the live analysis and captured alerts.
02 / From model to application
Connect your models. Configure your rules.
Configure model inputs, detection regions and event rules with built-in video processing, live views and integration APIs.
Visual scenario orchestration
Connect detection, tracking, filtering and rules. Adjust regions and logic as your requirements evolve.
Explore the pipeline editor →Detect→Track→Filter→Evaluate
Logic overview · View the full pipeline ↗PROMPT-BASED EXPLORATION
Describe a target or ask about a scene.
Use DINO to find targets described in a prompt, or ask VLM questions about a scene. Try your own images or video and inspect the detections, answers and events.
View platform benchmarks →Example: DINO with the garbage.person prompt and a corridor result.03 / SYSTEM ARCHITECTURE
Analyze at the edge.
Deliver events to your systems.
Video processing, inference and scenario rules run on the edge device. A web console handles configuration and inspection; interfaces deliver events to business systems.
On-site inputs
RTSP streams
Video & images
Web Console
Configure tasks · Live views · Review events
CosmoEdge Edge AI engineOn the edge device
Runtime platformsSophon / Rockchip / x86
Business systems
Receive events & results
Select a module to explore its role
Scenario orchestration
Connect models and processing nodes, define regions and set conditions. Save task logic as a scenario template, ready to adjust and reuse in the next task.
04 / ENGINEERING EVIDENCE
Performance and sustained operation. Tested separately.
Person detection + no-safety-helmet analysis · 2 tasks / stream · 5 FPS target per task
Multi-stream, dual-task performance
Highest tested passing workload
≥ means the highest tested workload passed.Passing gate: 80% of target FPS · 30 seconds per step
Read the performance report72 hours at a fixed load
72hContinuous operation
0%Maximum observed discard rate
| Platform | Fixed streams | Average FPS |
|---|---|---|
| BM1688 | 8 | 5.086 |
| CV186X | 8 | 5.085 |
| RK3576 | 8 | 5.098 |
| RV1126B | 4 | 5.230 |
A separate fixed-load run targeting 5 FPS per task. Average FPS comes from runtime samples.
Read the 72-hour reportTest input1080p · H.264 High · 24 FPS local looped video, with preview off.
05 / GET STARTED
Choose the right setup for you.
Deploy on your hardware
Run your first detection on a computer, then choose the build and deployment path for your edge hardware.
Choose a setup path →Start with included algorithms
Ready devices include vision models and scenario presets. Compare RK3576, CV186AH and BM1688. For international purchases, contact us by email.
Compare configurations →Discuss your deployment
Bring your model, hardware context and business needs to define integration scope and validation steps.
Discuss your project →BUILD YOUR FIRST FIELD APPLICATION
Your next step: get a task running.
CosmoEdgeApache-2.0 licensed enginev1.1.0