> ## Documentation Index
> Fetch the complete documentation index at: https://framework.freysa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Silo Box

> Your personal AI infrastructure at home

A workstation-class personal server designed for individuals who want to run large AI models privately at home. Your exocortex. Private, local, and entirely yours.

[Buy Silo Box](https://buy.stripe.com/8x228q4Y18t74YA0rygUM04)

## Who is this for?

For those who want their own Jarvis at home. For people who want no cloud dependency at all.

### All your data, one place

Connect all your personal data sources into a single, private system. Your emails, documents, photos, notes, calendars, and more. All indexed, all searchable, all under your control. A personal AI that works around the clock without sending a single byte to external providers.

### We build with you

This isn't just hardware. We work closely with Silo Box owners to build custom workflows that unlock the full potential of your data. Want to connect a new data source? Need a specific automation? We'll build it with you. Our goal is to continually expand what your personal AI can do.

## Core Specifications

### Graphics: 3x NVIDIA RTX 6000 Pro (Blackwell Max-Q)

* 96 GB GDDR7 each, 288 GB total
* PCIe 5.0 x16 per card
* 300 W TDP per GPU

### Processor: AMD Threadripper PRO 9975WX

* 32 Cores / 64 Threads
* 4.0 GHz base / 5.4 GHz boost
* 128 MB L3 Cache

### Memory: 512 GB DDR5 ECC

* 8 x 64 GB modules
* DDR5-5600 MT/s speed
* ECC for data integrity

### Storage: 2 TB NVMe + Expandable

* 1 x 2 TB PCIe 4.0 NVMe
* Optional 8-16 TB expansion
* ZFS/btrfs mirror support

## Complete Specifications

| Component        | Detail                                            |
| ---------------- | ------------------------------------------------- |
| GPU Model        | 3x NVIDIA RTX 6000 Pro Blackwell Max-Q            |
| GPU RAM          | 96 GB GDDR7 per GPU (288 GB total)                |
| GPU Interconnect | PCIe 5.0 x16 per card (\~64 GB/s each direction)  |
| CPU              | AMD Threadripper PRO 9975WX, 32C/64T, 4.0-5.4 GHz |
| System RAM       | 512 GB DDR5-5600 ECC REG (8 x 64 GB)              |
| Fast Storage     | 1 x 2 TB PCIe 4.0 NVMe SSD                        |
| Motherboard      | ASUS Pro WS WRX90E-SAGE SE                        |
| Networking       | Dual 10 GbE onboard; optional 100 GbE NIC         |
| Power Supply     | 2000 W 80 PLUS PSU                                |
| Power Draw       | \~1515 W max on 120 V / 20 A circuit              |
| Chassis          | Fractal Meshify 2 XL                              |
| CPU Cooling      | 360 mm AIO liquid cooler                          |

## What you can do with it

* **Run large LLMs locally**: Run the largest language models without quantization. PCIe-sharded inference across 3 GPUs.
* **Host multimodal agents**: Run multimodal models and autonomous agents privately. Vision, audio, and text. All local.
* **Personal cloud storage**: Replace Dropbox with mirrored local storage. Your data stays safe and entirely yours.
* **Fine-tuning and training**: Optimized for fast multi-GPU inference and fine-tuning on your own data.

## Home Deployment

Designed for real-world home use with practical power, thermal, and acoustic considerations.

* **Power**: \~1.5 kW typical load at 120V. Requires dedicated 15A or 20A circuit.
* **Thermals**: \~5,170 BTU/hr at typical load. Ensure clear exhaust path.
* **Acoustics**: \~45-65 dBA sustained. Quieter with power capping.

### Monthly Operating Cost

Estimated electricity cost for 24/7 operation:

| Mode               | Cost               |
| ------------------ | ------------------ |
| Home Mode (1.5 kW) | $162 to $270/month |
| Full Performance   | $173 to $288/month |

[Buy Silo Box](https://buy.stripe.com/8x228q4Y18t74YA0rygUM04)