Ultralytics YOLOv5, a PyTorch-based computer vision toolkit for object detection, instance segmentation and image classification with training and export.
58.2K17.5KPythonAGPL-3.0Updated 2 days ago
cosmo-wander-ai/cosmo-edge
C++ edge AI engine that turns video analytics and on-device VLM models into deployable applications on Sophon, Rockchip RKNN, and x86 hardware.
- Stars
- 1,838
- Forks
- 312
- Open issues
- 21
- Last commit
- yesterday
What is it?
CosmoEdge is an edge AI engine written in C++ for running video analytics and vision-language model (VLM) workloads on edge devices. It covers model import, visual pipeline orchestration in a browser console, alarm rules, OSD overlays, and event delivery over REST, WebSocket, MQTT, and HTTP webhooks.
Each build targets one inference backend: BMRT for Sophon BM1688 and CV186X, RKNN with MPP/RGA for Rockchip RK3576 and RV1126B, and ONNX Runtime with FFmpeg on x86 Linux and Windows. Models must be generated for the chip they run on. An x86 mode runs the same console and workflow without edge hardware, at lower throughput.
The project includes GroundingDINO open-vocabulary detection, prompt-based VLM judgment, ONVIF and GB28181 camera access, and a Docker-based build for each platform. Its v1.1.0 release includes published benchmark reports for each supported chip.
Key features
- 01Visual pipeline composition and task binding in a browser console.
- 02Detection, classification, tracking, zones, counters, OSD, and alarm snapshots.
- 03Prompt-based VLM judgment and GroundingDINO open-vocabulary detection.
- 04Event delivery over REST, WebSocket, MQTT, and HTTP webhooks.
- 05ONVIF and GB28181 camera access.
- 06Docker-based build paths for Sophon, Rockchip, and x86.
Use cases
- Running multi-channel video analytics for safety or intrusion detection on edge boxes.
- Adding prompt-driven visual checks that report alarms to an existing system over MQTT or webhooks.
- Prototyping a video AI workflow on x86 before deploying to Sophon or Rockchip hardware.
Quick start
- 1
Clone the repository
git clone https://github.com/cosmo-wander-ai/cosmo-edge.git - 2
Enter the project directory
cd cosmo-edge - 3
Start the x86 Linux version with Docker Compose
sudo docker compose -f docker-compose.x86.yml up -d --build - 4
Open the web console at localhost:8080
Things to know
- Edge NPU deployments require specific hardware (Sophon, Rockchip); the x86 mode runs with lower throughput.
- Apple Silicon macOS support is only a Preview using amd64 emulation for a single local-video workflow.
- Current main branch changes are not included in the stable v1.1.0 packages.
Topics
FAQ
- What is cosmo-edge?
- C++ edge AI engine that turns video analytics and on-device VLM models into deployable applications on Sophon, Rockchip RKNN, and x86 hardware. CosmoEdge is an edge AI engine written in C++ for running video analytics and vision-language model (VLM) workloads on edge devices. It covers model import, visual pipeline orchestration in a browser console, alarm rules, OSD overlays, and event delivery over REST, WebSocket, MQTT, and HTTP webhooks.
- Is cosmo-edge free to use? What license does it have?
- cosmo-edge is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
- Is cosmo-edge still maintained?
- Actively maintained. The last commit to cosmo-edge was on Oct 19, 2026.
- What can I use cosmo-edge for?
- Running multi-channel video analytics for safety or intrusion detection on edge boxes. Adding prompt-driven visual checks that report alarms to an existing system over MQTT or webhooks. Prototyping a video AI workflow on x86 before deploying to Sophon or Rockchip hardware.
- How do I install cosmo-edge?
- Clone the repository: git clone https://github.com/cosmo-wander-ai/cosmo-edge.git Enter the project directory: cd cosmo-edge Start the x86 Linux version with Docker Compose: sudo docker compose -f docker-compose.x86.yml up -d --build Open the web console at localhost:8080
- Where can I get cosmo-edge?
- The source code and installation instructions are on GitHub: https://github.com/cosmo-wander-ai/cosmo-edge
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A tiny on-device foundation model for tool calling, structured extraction, embeddings and speech-to-text on phones, wearables, robots and microcontrollers.
14.1K1.1KPythonApache-2.0Updated 2 days ago
Open-source toolkit for optimizing and deploying deep learning models for inference on Intel CPUs, GPUs and NPUs.
11K3.5KC++Apache-2.0Updated 6 hours ago
A cross-platform desktop app for AI-assisted annotation of text, image, video, point cloud and multimodal data, with built-in models and multi-format export.
AI & Machine Learning#image-annotation#computer-vision#deep-learning#object-detection#instance-segmentation
10.7K1.2KPythonGPL-3.0Updated 2 days ago
PyTorch's on-device inference stack for running exported models on phones, laptops, browsers, and microcontrollers.
5.1K1.2KPythonOtherUpdated 4 hours ago
A C++ video analysis framework that chains stream reading, model inference, tracking and output as pluggable pipeline nodes.
3K470C++Apache-2.0Updated 8 months ago