Endless — Open Source Intelligent IP Camera Solution

Version 1.0 · March 2026

Product Overview

Endless is a production-grade, fully open-source intelligent IP camera system built on embedded Linux. It delivers a complete software stack — from SoC-level hardware abstraction to cloud AI integration — covering the full pipeline: video capture, multi-protocol stream distribution, local AI inference, multi-modal AI agent services, and a web-based management interface.

The project is designed for teams who need a solid, customizable firmware base rather than a black-box SDK, with support for hardware selection, board adaptation, peripheral integration, and software feature customization.

Target audiences:

Hardware Customization Overview

Endless is not limited to generic firmware delivery. It also supports hardware customization around the customer's target product form factor. For teams building commercial camera products, the existing platforms can be extended with board adaptation, peripheral integration, and product-specific configuration.

Customization Scope

Delivery Approach

• Start from existing reference platforms to shorten hardware bring-up time
• Create different variants for commercial, entry-level, or high-end AI camera products
• Run hardware customization and software tailoring in parallel to reduce integration cost

Core Capabilities

Video Pipeline

Multi-Protocol Stream Distribution

AI Intelligence

The Qwen3-Omni integration supports real-time audio-visual interaction — users can verbally query the camera ("Is there anyone at the entrance?") and receive spoken AI responses based on live video analysis.

Hardware & Software Customization

• Select from existing Hisilicon/QEMU platforms and extend to new SoCs or board profiles
• Customize peripherals such as GPIO, USB audio, IR-CUT, watchdog, and board-level drivers
• Tailor web UI, RESTful API, AI, storage, and protocol modules based on product scope
• Extend branding, business workflows, protocol adapters, and cloud integrations as needed

Industry Protocol Support

• ONVIF: Device discovery, media service, device management service
• GB28181: SIP signaling, PS stream encapsulation, platform registration
• RESTful API: Full HTTP/HTTPS API covering auth, config, media, storage, and upgrade
• WebSocket: Real-time event push (AI detections, system alerts, telemetry)

Storage

• microSD card with hot-plug support and filesystem monitoring
• Continuous recording with timestamp indexing
• Configurable loop-overwrite policy with storage-space alerting
• Timeline-based playback and recording search via web UI
• Download via API or web interface

Security

• HTTPS with TLS on all web access
• Challenge-response authentication (anti brute-force)
• JWT token session management with automatic refresh
• Data encryption for sensitive configuration
• Firmware signing and verified boot support (hardware-dependent)

System Architecture

┌─────────────────────────────────────────────────────────────────┐
│  Application Layer                                               │
│  Web UI (Vue 3) · RESTful API · ONVIF · GB28181 · AI Agent UI  │
├─────────────────────────────────────────────────────────────────┤
│  Foundation Services                                             │
│  Stream Service (RTSP/WebRTC/RTMP) · Agent · Qwen · Xiaozhi    │
├─────────────────────────────────────────────────────────────────┤
│  Base Business Layer                                             │
│  Media · Record · Storage · File Mgr · Upgrade · Rule Engine    │
├─────────────────────────────────────────────────────────────────┤
│  Base Software Layer                                             │
│  Config Mgr · Event Bus · GStreamer · FFmpeg · SQLite · OpenSSL │
├─────────────────────────────────────────────────────────────────┤
│  HAL (Hardware Abstraction Layer) — pure C interface             │
│  Video · Audio · AI Accelerator · Peripheral · USB              │
├─────────────────────────────────────────────────────────────────┤
│  Kernel Layer: Linux 5.10 + SoC device drivers                  │
├─────────────────────────────────────────────────────────────────┤
│  Hardware: Hisilicon 3516CV610 · 3516CV608 · 3519DV500 · QEMU  │
└─────────────────────────────────────────────────────────────────┘

See Architecture Document for full layer descriptions, component relationships, and data flow diagrams.

Hardware Platform Support

Cross-compilation is handled by platform-specific toolchains under device/soc/hisilicon/. The QEMU platform enables full-stack development and regression testing without physical hardware, making software customization verification faster.

Technology Stack

Backend

Frontend

AI Services

Typical Use Cases

Home & SMB Security

Deploy on Hisilicon 3516CV610 hardware. Connect via browser (WebRTC live view < 500 ms latency). AI detects persons and vehicles locally; events are stored to SD card. Users can ask the camera voice questions via the AI Agent interface.

Enterprise Video Intelligence

Register cameras to existing VMS via ONVIF or GB28181. Use the RESTful API to integrate with custom dashboards or SIEM systems. Enable cloud Qwen-VL for advanced scene description and anomaly analysis.

OEM / ODM Product Development

Use Endless as a firmware baseline. The HAL abstraction isolates hardware-specific code, making SoC transitions, board bring-up, and peripheral adaptation manageable. C++ service modules are independently compilable, and web, protocol, and AI capabilities can be tailored to product scope. Custom AI pipelines can be added as separate modules under foundation/.

Edge AI Research & Development

The QEMU aarch64 platform (Product PB) provides a complete functional environment for algorithm development, CI testing, and integration work without physical hardware. YOLO model uploads and hot-reload are supported via the web interface.

Integration Reference

API Endpoints (excerpt)

WebSocket Events

RTSP Stream URL Format

rtsp://<device-ip>:<port>/stream/<channel>

Default port: 554. Supports up to 5 concurrent RTSP connections.

Roadmap

Near-term

• WebRTC end-to-end encryption
• Expanded AI detection scenarios and custom model management
• AI inference performance optimization (quantization, NPU scheduling)

Mid-term

• Cloud platform APP integration
• Mobile APP (iOS / Android)
• Developer SDK and documentation portal

Long-term

• Additional hardware platforms and customization reference designs
• Edge AI model optimization (INT8, pruning)
• Multi-camera coordination

Project Links

• GitHub: https://github.com/endless-sky-bee1/endless
• Issues: https://github.com/endless-sky-bee1/endless/issues
• Email: endless@endless-sky.onaliyun.com