Neo Grid: serve models across your computers

Your office computers, workstations and servers serve your models together, behind one private API that sends each request to the least busy one.

Create inference

• Your computers
• Each joins with one command
• Neo Grid
• One private API
• Apps
• API
• Harness
• Your team
• Detail A · Routing
• Each request to the least busy computer
• Busy
• Free
• Busy
• Off

1. Create

   Create your inference: your address, admin panel and API.

2. Connect

   Each computer joins with one command.

   yougori inference connect URL

3. Approve

   Choose what each computer gives, then approve it on the website.

4. Route

   Each request goes to the least busy computer.

5. Use

   Your team calls your API from its apps, tools and harnesses.

Turn a computer off. Nothing stops.

Switch off any of these computers. Its requests move to the others, and your team sees no error.

• Design workstation
• Server room 1
• Reception desktop
• Studio PC
• Your team
• Requests answered
• Errors your team saw
• Three tries
• Up to three computers try a request before anything reaches your app.
• Eight seconds
• A request no computer takes within 8 seconds goes to another one.
• Thirty seconds
• A computer that fails goes to the back of the line for 30 seconds.

Every computer in one view.

Your admin panel shows what each computer serves, how busy it is and how its GPUs are doing. Example figures:

• Routing
• Model
• Served by
• In progress / slots
• Requests
• Same conversation
• Failed
• State
• Chat model
• Design workstation
• 3 / 8
• 1,284
• 862
• 0
• Serving
• Chat model
• Server room 1
• 5 / 8
• 2,041
• 1,377
• 2
• Serving
• Coding model
• Studio PC
• 1 / 8
• 416
• 301
• 0
• Serving
• Image model
• Reception desktop
• 0 / 8
• 58
• 0
• 1
• Skipped after a failure
• GPUs
• GPU
• Utilization
• VRAM used / total
• Temperature
• Power
• Design workstation
• 64%
• 18 / 24 GB
• 61 °C
• 248 W
• Server room 1
• 88%
• 41 / 48 GB
• 70 °C
• 281 W
• Studio PC
• 23%
• 9 / 16 GB
• 48 °C
• 96 W

Each computer chooses what it shares.

When a computer joins, it picks the models it serves and the room it lends. Everything else on it stays private.

• Studio PC
• Shared
• Private
• Chat model
• Coding model
• Image model
• Sandbox VMs
• 8 CPU cores · 16 GB · 2 slots
• Files
• Code 7KQ2MX4P
• Approved on your website
• Your inference
• Up to 8 at once
• Each computer answers up to 8 requests at a time.
• Approved on your website
• A computer serves only after its code is approved.
• Leave any time
• Disconnect it on the website, or stop it on the computer.

Questions.

What is Neo Grid?

Neo Grid serves AI models across your own computers behind one private API. Office desktops, workstations and servers each run models, and your inference sends every request to one of them.

Which computers can join?

Computers you own with Yougori installed: office desktops, workstations and servers. Each one chooses the models or sandbox capacity it gives.

How are requests shared?

A conversation stays on the computer that started it, so its cached input is reused. Other requests go to the least busy computer, and if one fails before it answers, another takes the request.

Can rented GPUs join too?

Yes. GPUs you rent join the same inference as your own computers, for the hours you need them.

Do our models stay on our computers?

Yes. The models run on the computers you connect. Your inference passes the requests and replies.

What does it cost?

Creating an inference on yougori.com is free while it is in early access. You pay your GPU provider directly for any GPUs you rent. Our help to plan, set up and speed up your inference is priced per project.

YOUR COMPUTERS. YOUR INFERENCE.

Put your computers to work.