Edge AI Meets Network Storage

The most powerful AI NAS for real edge computing, AI workflows, and total data sovereignty

Full size PCIe and graphics card compatible
Dual 10GbE
Up to 96GB ECC memory
Up to 132GB Storage: 4 HDD + 3 SSD

Full size PCIe and graphics card compatible
Dual 10GbE
Up to 96GB ECC memory
Up to 192GB Storage: 4 HDD + 3 SSD

*NOT ALL Software features are ready on day 1.

Ask your files anything

Private answers with source citations—on your device.

Find the right shot, fast

Describe it and get tagged results and transcripts—locally.

Ingest. Edit. Export—without copying

SD/USB4 ingest and 10GbE editing straight from the NAS.

Agents that do the busywork

Scheduled summaries, smart routing, and file actions.

Your morning brief, from your data

Daily digests from folders, inboxes, and drives—no cloud needed.

Private by default. Works offline.

On‑device AI. No subscriptions. Your data never leaves your N1.

Your private AI stack, on Docker

Run agent frameworks, host A2A and MCP servers; add a graphics card for acceleration.

For builders:
your private AI stack

Container-native platform for AI agents, LLMs, and custom services—locally hosted, fully yours.

Local AI suite
(multimodal, private, on your device)

Runs on Nexus AI Station by default. No cloud unless you opt-in.

Multimodal RAG

Text, images, audio/video, structured & semi-structured data

Extract, embed, chunk; link metadata across sources

Hybrid index & search (keyword + vector)

On-device by default; citations included

Natural‑language Console

Type or speak; get answers, links, media snippets

Toggle local models (default) with optional cloud models

Long‑context support; improves with your usage history

Trigger agents with commands (e.g., /summarize, /route)

Multimodal RPA

File governance: classify, rename, de‑duplicate and identify files with local LLMs.

Capture & extract from PDFs, images, audio/video

Smart email management: summarize, label, route

Schedules, rules, and audit logs

Research & Development

Train cutting-edge models, experiment with new architectures, and push the boundaries of AI research with unprecedented computational power.

Large Language Model Training
Computer Vision Research
Reinforcement Learning

Production Deployment

Deploy AI models at scale with enterprise-grade reliability, monitoring, and security features built for mission-critical applications.

Model Serving & Inference
Real-time Processing
Edge Deployment

AI Agent Development

Build and deploy sophisticated AI agents with our integrated development environment and orchestration platform.

Multi-Agent Systems
Workflow Automation
Custom Agent Frameworks

Real AI needs real GPU power

Nexus is built to run local AI faster and easier—just add your graphics card.

Bring your own GPU Card up to RTX 5070\n(not included in the box)

Supports full-size NVIDIA GeForce RTX cards up to RTX 5070 (16GB)

PCIe Gen4 x16 slot (electrical x8 lane) with full-length, full-height clearance

800W internal PSU provides headroom for AI workloads

Harbor OS detects/configures the GPU card for AI apps (RAG, vision, agents)

Performance depends on model/VRAM and workload. Large models may require

Performance you
can trust

AMD Ryzen 7 PRO 8845HS with dedicated NPU for smooth AI and everyday workloads, born for 24/7 devices. Up to 38 TOPS total AI performance.

8-core Ryzen 7 PRO

Efficient always-on design

Great for containers/services

ECC RAM for data integrity

We publish real-world benchmarks before shipping (local LLMs, RAG, ingest/edit.)

Up to 132GB

Capacity you can grow

4 × 3.5" HDD for your library

3 × M.2 NVMe for cache and active projects

Enterprise-level data protection with ZFS and ECC RAM

Access your files and agents securely from any device.

Auto-backup files (iOS/Android)

Share albums and files with a link

Safe mobile backup and browse

Secure access

for teams

Sign-in controls and permissions

Optional remote access - local AI still runs on the device

Use your phone or laptop to reach your files, photos, and agents—on your terms.

Dual 10GbE

2,000+ MB/s

1GbE NAS

125 MB/s

Public cloud

30–50 MB/s

Indicative targets; results vary by drives, network, cache

Full Function
USB4 and SD card slot

UHS-II SD card slot, over 300MB/s Direct-attach at up to 40Gbps

PCIe expansion for your GPU card

Full-size discrete GPU card support

PCIe Gen4 x16 (electrical x8 lane) with 800W PSU headroom

Built by the Community, For the Community

Harbor OS is completely open source – your AI workstation, your rules

100% Open Source

Full access to Harbor OS source code on GitHub. Transparent, auditable, and community-driven development.

Community -Driven

Feature requests, bug fixes, and improvements come from real users. Your voice shapes the roadmap.

No Vendor Lock-in

Install alternative OS if desired (Unraid, TrueNAS, Proxmox). Your hardware, your choice.

Build With Us

Developer-Friendly Ecosystem

APIs & Documentation

Comprehensive APIs and documentation for custom integrations and automation.

Transparent Roadmap

Community votes on priority features and development direction.

Join the Build

Contribute code, test features, and shape the future of private AI computing.

Join Our GitHub Community

Feature

Nexus AI Station

Synology DS923+

UGREEN NASync

DIY Unraid

AI Support

Local AI Compute & Storage

Storage & Backup

Local AI & Storage

Flexible Server Hosting

GPU Support

Full-Size dGPU (RTX 5070)

None

eGPU via OCuLink

Yes, with limitations

PSU

800W Internal

100W External

External (TBD)

User-Supplied

Local LLM Ready

Yes (Integrated OS)

No (Cloud API only)

Yes (Beta Software)

Yes (Complex Setup)

Software Setup

High (Launch Target)

High

Low (Beta)

N/A (User-Managed)

10GbE LAN

Yes

Optional Upgrade

Yes

User-Supplied

Setup Simplicity

High

Medium

Low

Noise Level

Whisper-quiet (3 thermal modes)

Moderate

Variable

User-dependent

Open Source OS

Yes (Harbor OS)

No (Proprietary DSM)

No (Proprietary UGOS)

Yes (Multiple options)

Community Development

GitHub-driven roadmap

Closed development

Community-driven

800W

PSU

full-size

GPU Card

PCIe

Gen4 x16 (x8 lanes)

4 HDD +

3 NVMe tiers

3.0"

touch LCD

Quiet

16L

zoned cooling

Dual 10GbE

Network Ports

Up to 96GB

ECC

DDR5-5600

HDMI 2.0

Full Function USB4

UHSII SD card slot

Barebone

No RAM, No Storage Bring your own components

$799

Base system only

Get Early Access

Starter Kit

32GB RAM + 512GB NVMe Ready for AI workloads

$999

Best for homelab

Get Early Access

Join the waitlist and get:

Upto200off

Early-bird pricing
First 25 get a free 512GB NVMe cache drive upgrade
30-day returns and 3-year warranty

Install drives

Install drives (and your graphics card if you have one)

Power on

Power on and complete setup in Harbor OS

Finish configurations

Click to install AI tools (chat over docs, media intelligence, agents)

Detailed Specifications

Compute

CPU: AMD Ryzen 7 PRO 8845HS (8-core, 16-thread)

Base Clock: 3.8 GHz

Boost Clock: Up to 5.1 GHz

Architecture: Zen 4

TDP: up to 55W

Memory

Type: DDR5-5600 So-dimm

Slots: 2 × SO-DIMM

Capacity: 96GB ECC (2 × 48GB)

Default: 32GB (2 × 16GB)

Expansion/GPU card

PCIe Slot: 1 × PCIe 4.0 x16 (x8 lane)

GPU Support: Up to RTX 5070 Ti / RX 7800 XT

Power Budget: 800W total system power

Clearance: 335mm length, Full size

Storage

HDD Bays: 4 × 3.5" SATA III (hot-swappable)

NVMe Slots: 3 × M.2 2280 PCIe 4.0

RAID Support: 0, 1, 5, 6, 10, JBOD

Networking

Ethernet: 2 × 10GbE (RJ45)

Controller: Marvell AQC-113C

Link Aggregation: Supported

Wake-on-LAN: Supported

I/O

USB4: 1 × USB4 40Gbps (Type-C)

USB-A: 1 USB 3.2 Gen2(10Gbps), 1USB 3.2 Gen1 (5Gbps), 2x USB 2.0

SD Card: 1 × UHS-II (up to 312 MB/s)

HDMI: 1 × HDMI 2.0 (4k@60Hz)

System & Display

Display: 3.0" touchscreen (170 x 560)

Buttons: Power

LEDs: Status, Network, HDD activity

Sensors: Temperature, fan speed

Dimensions & Build

Form Factor: custom-designed board

Dimensions: 218x193x372mm(L×W×H)

Material: Aluminum chassis, steel frame