Now in development · Est. 2026

Your AI,
in your pocket.

The world's first pocket AI box. Plug into any laptop,run 14B models locally. No cloud. Total privacy.

A minimalist white box, 91×91×31mm — about the size of an Apple TV. Rounded corners, matte finish, completely silent. Two USB-C ports on the side: one for power, one for data. A single LED glows softly when your models are running.

91×91×31mm280gWhite matte
0 TOPSNPU Power
0GBMemory
Up to 0BModel Size
~0gWeight

The Problem

Thinlaptopsweren'tbuiltforAI.

💻

Not enough RAM

Most ultrabooks have 8–16GB RAM shared with the GPU. Running a 7B model leaves nothing for your apps.

🐌

No GPU power

Integrated graphics struggle. A 7B model on CPU runs at 2–5 tok/s — unbearably slow for real work.

☁️

Cloud = no privacy

Sending code, documents, and conversations to cloud APIs means you don't control your data.

The Solution

Thinkofitasapowerbank.ForAI.

Plug into any USB-C port. Your laptop instantly gets 45 TOPS of dedicated NPU and 16GB of dedicated AI memory. Open a browser tab at localhost:8080 and chat with your models. No setup. No cloud. Just works.

Zero configUSB Gadget modeOllama + Open WebUI16GB LPDDR5
🔌
One cable. Infinite AI.

How It Works

Plugin.Openbrowser.Askanything.

🔌
Step 1

Connect

Plug the AI Box into your laptop via USB-C. It powers up instantly — no wall adapter needed.

🌐
Step 2

Open Browser

Navigate to localhost:8080. Open WebUI loads automatically with Ollama pre-configured.

💬
Step 3

Start Chatting

Download any model from the built-in market. Qwen, DeepSeek, Llama — all run locally, privately.

💻Your Laptop
USB-C
📦AI Box
local
🧠Private AI

Use Cases

AIthatgoeswhereyougo.

💼

Office & Remote Work

Run code assistants, meeting transcription, and document RAG without sending company data anywhere.

🎨

Creative Design

Generate images with Stable Diffusion locally. Iterate faster with no API rate limits.

🎓

Education

Take your personal AI tutor anywhere. Offline-capable, no WiFi needed after model download.

✈️

Travel

280g fits in any bag. On flights, in cafes, at hotels — your AI is always with you.

🔒

Privacy-First Work

Lawyers, doctors, journalists — handle sensitive data with local AI. Nothing leaves the box.

🛠️

Developers

Local LLM code review, test generation, and documentation. Integrates with VS Code via Continue.dev.

Performance

Actuallyfast.Notcloudfast.

Real-world INT4 speeds on CIX P1 NPU

Qwen 2.5 7B
30–40 tok/s
82%
Qwen 2.5 14B
15–20 tok/s
55%
DeepSeek-R1-Distill 14B
12–18 tok/s
48%
Llama 3.1 8B
25–35 tok/s
72%
Whisper Large v3
2× realtime
65%

Technical Specs

Everythingyouneed.Nothingyoudon't.

Chipset
CIX P1 12-core
Architecture
ARM v9.2-A
NPU
45 TOPS INT8
GPU
IMG BXM-8-256
Memory
16GB LPDDR5-6400
Storage
512GB NVMe SSD
Expandable
M.2 2280 slot
Connect
USB-C 3.2 Gen 2 ×2
Network
Wi-Fi 6 + BT 5.3
USB Gadget
Virtual Ethernet
Battery
5,000mAh Li-Po
Bus Power
≤15W via USB-C
Dimensions
91×91×31mm
Material
Aluminum + ABS
Weight
~280g
Cooling
Passive (silent)
Noise
0 dB
OS
Debian Linux
AI Stack
Ollama + Open WebUI
Models
Qwen 2.5 / 3, Llama 3
Quantization
GGUF INT4 / INT8

Why AI Box

Therearefastercomputers.Butnonelikethis.

The Insight

AI doesn't need
a whole computer.

A mini PC runs an entire operating system — Windows, drivers, background services, antivirus, updates. 90% of its compute goes to things you don't need for AI.

AI Box runs one thing: your models. Every transistor, every watt, every byte of memory is dedicated to inference. That's why it punches above its weight.

OS + Background (82%)
AI (18%)
AI (100%)
Mini PC
AI Box

Head to Head

6 things AI Box does that mini PCs can't.

🔋

Runs on battery

5,000mAh built-in. Use it on a plane, in a cafe, or on the couch. Mini PCs are chained to a wall outlet.

📏

Fits in your pocket

91×91×31mm, 280g. Smaller than a wallet. A mini PC needs a backpack and a power brick.

One cable. That's it.

USB-C carries power + data + networking simultaneously. No HDMI, no Ethernet, no power adapter.

🤫

Completely silent

Passive cooling. Zero moving parts. Mini PCs have fans that spin up under AI workloads.

🧠

Dedicated NPU, not shared iGPU

45 TOPS on a purpose-built NPU vs. ~8 TOPS on a shared iGPU that also runs your display.

📦

Pre-installed AI stack

Ollama + Open WebUI out of the box. Mini PCs need hours of Linux setup, driver config, Docker.

Value

More AI power per dollar.

Best Value
AI Box
¥2,699
one-time
45 TOPS
51 ¥/TOPS
零刻 SER8
¥2,599
one-time
8 TOPS
325 ¥/TOPS
Mac Mini M4
¥4,799
one-time
38 TOPS
126 ¥/TOPS

The Hidden Math

Cloud API costs add up fast.

UsageCloud API / yearAI Box
Casual (20 prompts/day)~$240/yr¥2,699 once
Power user (100 prompts/day)~$960/yr¥2,699 once
Developer (500 prompts/day)~$4,800/yr¥2,699 once
以上基于 ChatGPT Plus $20/月估算。重度使用或 API 按 token 计费会更贵。
🎯

The honest truth.

AI Box is not trying to replace your computer. It's for people who already have a laptop they love — and just want it to run AI. If you need a new desktop, buy a mini PC. If you need your existing laptop to think faster, get an AI Box.

FAQ

Youask.Weanswer.

What models can it run?

Any GGUF model up to ~14B parameters at INT4. This includes Qwen 2.5, Llama 3.1, DeepSeek-R1-Distill, and Whisper for speech-to-text. Larger models may swap but we recommend 7B–14B for optimal speed.

How is this different from a mini PC?

Size (pocket vs. desktop), power (bus-powered vs. wall), and purpose. Mini PCs are general computers. AI Box is a dedicated AI accelerator — plug in and it just works, no OS setup needed.

Does it need internet?

Only for the initial model download. Once models are cached on the 512GB NVMe, everything runs fully offline. Perfect for flights, secure facilities, or low-connectivity areas.

What laptops are compatible?

Any laptop with a USB-C port that supports data. Works on Windows, macOS, and Linux. We recommend USB-C 3.2 for best throughput, but USB 3.0 is sufficient.

How long does the battery last?

The 5,000mAh battery provides ~4–6 hours of continuous inference. When plugged into a laptop, it draws power from the USB bus — no separate charger needed.

Be the first to know

Ready for local AI?

We're preparing the first production batch. Sign up for early access, launch updates, and an exclusive early-bird discount.

🔒 No spam. Just launch updates.

AI Box · Privacy-first computing · Est. 2026