One-click Deploy Agents and AI Applications
Quickly run multiple independent agents and AI applications on your Arm device
On your Arm device terminal
Core Features
AI application infrastructure built for Arm devices
One-Click Deploy
No manual environment setup. Jishu handles everything — get running in minutes
Capability Integration
Scan to connect WeChat, Feishu — quickly use various Skills, MCP and rich tools
Security Sandbox
Container-based isolation for Agent apps and system environment, independent API Key management, selective data authorization
Multi-Instance
Run multiple Agent instances in parallel on a single device, fully utilizing hardware capabilities
Unified Management
Centrally manage versions, configs, and status of all AI apps from a single interface
Why Not PC or Cloud?
Standalone Arm devices running Agents deliver the best experience
- Independence: PCs serve multiple purposes — work and life data interfere with each other
- Environment: Complex Windows desktop limits Agent capabilities, must stay awake
- Power: 5W vs 100W+ (Raspberry Pi 5 ~$70, PC costs hundreds) — significant difference for 24/7 operation
- Security: All data uploaded to cloud — leak and abuse risks
- Design: Contradicts standalone AI principle of local data, local execution — not private deployment
- Experience: Higher network latency, IPs flagged as server not personal — web access restricted
- As local computing power improves, data interaction and Agent execution can be powered by local GPU/NPU, enabling fully offline operation and ensuring data privacy
- Connect cameras, wheels, limbs, and sensors to turn your Arm device into a truly perceptive, actionable embodied AI terminal
Supported Hardware
Hardware baseline: Raspberry Pi 4, 4-core Arm Cortex-A72 or above, 4GB RAM, 16GB storage OS: Ubuntu 22.04+, Debian 12+, MacOS 26
Raspberry Pi 5/4
BCM2712/BCM2711 · 4×Cortex-A76/A72 · 4/8GB
Affordable, mature ecosystem — ideal for running standalone AI apps
Minimum: Cortex-A76/A72 quad-core or above + 4GB RAM
Nvidia Jetson Orin / Thor
Cortex-A78AE / Neoverse-V3AE · 4~128GB
Native Nvidia GPU integration — supports local model inference at varying scales
Ref: Jetson Orin Nano · Jetson AGX Orin · Jetson Thor T4000/T5000
Mac Mini / Studio
Apple Silicon M1~M5 · Unified Memory 8~192GB
Apple unified memory architecture — peak local inference, top choice for multi-instance on a single device
Rockchip RK3588/RK3576
4×Cortex-A76/A72 + 4×Cortex-A55 · up to 32GB
Cost-effective domestic edge chip, 6 TOPS NPU, octa-core big.LITTLE architecture
Ref board: Firefly EC-I3588J · Forlinx FET3588-C · MYIR MYD-LR3576
CIX P1
8×Cortex-A720 + 4×Cortex-A520 · up to 64GB
Domestic high-performance Arm SoC, ARMv9.2 cores, with 30 TOPS "Zhouyi" NPU
Ref board: Radxa Orion O6
Huixi Guangzhi R1
24×Cortex-A78AE · 32~128GB
Domestic automotive-grade high-performance AI chip, 500 TOPS Rhino NPU
Mediatek Genio720
2×Cortex-A78 + 6×Cortex-A55 · up to 16GB
Leading AIoT chip, 6nm process, 10 TOPS NPU, Antutu score up to 800K
Ref board: Zelustek G720 development board
More Arm SoCs under evaluation — other chips and devices are welcome to try, stay tuned for announcements
Coming SoonScan to Join the Community
Join our community group for the latest product updates
微信扫码