Pre-order / Contact
New from DoAI

MindBox Mini

The AI Token Machine — on your desk.

A silent, fanless desktop computer built to run private, agentic AI locally: no cloud, no data leaving your office.

64 TOPS NPU 64GB LPDDR5 < 36W 7B LLM @ 20 T/s Fanless & silent
Pre-order / Contact View full specification
MindBox Mini — compact fanless desktop AI machine in silver, front view with perforated grille and power button
MindBox Mini · 200 × 210 × 85 mm · fanless
Why MindBox Mini

Local agentic AI, without the cloud

Three reasons it earns a permanent spot on your desk.

Truly silent & desk-friendly

Designed to disappear into your workspace — no fan, no noise, no footprint you’ll notice.

  • Fanless & ultra quiet
    zero fan noise for a calm office or personal desk
  • Ultra compact
    200 × 210 × 85 mm of serious compute

Power-efficient local agentic AI

Run around the clock for pennies, and keep every token on your own hardware.

  • < 36W draw
    dramatically cheaper to run 24/7 than a GPU workstation
  • OpenAI-compatible API + 64 TOPS NPU
    deploy local agents & workflows without the cloud

Robust AI workflow capabilities

Enough muscle for everyday models, plus the bandwidth to keep complex pipelines smooth.

  • Small-LLM ready
    runs 7B LLMs at ~20 T/s for reasonable everyday AI tasks
  • 64GB LPDDR5 unified memory + 8-core A55
    for smooth data handling on complex models
  • Expandable & connectable
    up to 2TB M.2 SSD (PCIe Gen4) + dual 1Gbps Ethernet
Specification

Technical details

Compute & Memory

AI Processor
SOPHGO CV84X6
Neural Engine
64 TOPS @ INT8 / FP8 NPU
CPU
8-core Arm Cortex-A55
Memory
64GB LPDDR5 high-bandwidth unified memory
Video decode
H.264 / H.265, 20 T/s codec;
64 simultaneous streams (2 MP @ 30fps)
LLM performance
7B small LLM models at ~20 T/s

Form Factor, Power & I/O

Dimensions
200 × 210 × 85 mm (ultra compact)
Power
< 36 W, super power-efficient
Cooling / noise
Fanless, ultra quiet
Storage
Up to 2TB M.2 SSD, PCIe Gen4
Networking
Dual 1 Gbps Ethernet
Display output
HDMI 2.0

AI Platform & Software

API
OpenAI-compatible API — drop in existing clients and tools
Agentic AI
Ready — run local agents and multi-step workflows entirely on-device
Deployment
Fully local & private — your data never has to leave the building
Q&A

Good to know

What is MindBox Mini?
MindBox Mini is a desktop “AI Token Machine” from DoAI — a silent, fanless computer designed to run local LLMs and agentic AI workflows on your own hardware, using the SOPHGO CV84X6 with a 64 TOPS NPU and 64GB of unified memory.
How is it different from cloud AI?
Everything runs locally. Your prompts, data, and models stay in your office — no cloud round-trips, no per-token service fees, and no dependence on external APIs. It exposes an OpenAI-compatible API, so tools you already use work as-is.
Can I run useful models on it?
Yes. The 64GB unified memory and 64 TOPS NPU comfortably host 7B-class small LLMs at roughly 20 tokens per second — well suited to everyday assistant, RAG, and agentic tasks. It also handles video decode (up to 64 concurrent 2MP/30fps streams).
How loud and how much power does it use?
It’s fanless and ultra quiet, and draws under 36W — making it very cheap to keep running 24/7 compared with a traditional GPU workstation.
Is it expandable and connectable?
Yes. It supports up to a 2TB M.2 SSD (PCIe Gen4) for local storage, dual 1Gbps Ethernet for reliable networking, and HDMI 2.0 for display output.
Get MindBox Mini

Put an AI Token Machine on your desk

Interested in pre-orders, enterprise rollout, or developer access? Reach out and we’ll point you to the right next step.

WhatsApp us [email protected]