A partner to help you create inference

We build your company’s own AI inference with you, on your computers or GPUs you rent, guaranteed 1.5× faster with our own custom megakernels.

Create it yourself

Your inference

Your own address · admin panel · API

Your computers

In your office

Rented GPUs

By the hour

Detail A · Megakernel

The whole model in one GPU launch

• Harness
• Video
• JSON
• Images
• Audio

Others

Yougori

1.5×

At least 1.5× faster on your GPU

or you do not pay for it

1. Plan

   Your models, users and budget decide the hardware.

2. Create

   Your website, admin panel and API at your own address.

3. Connect

   Each computer joins with one command.

   yougori inference connect

4. Speed up

   Custom megakernels for your model and GPU.

5. Hand over

   Your team connects its harness or any OpenAI-compatible tool.

Everything your inference comes with.

One build gives your company all of it, at your own address, on the computers you choose.

• Your own address
• acme.yougori.com, with your name, your words and your colour.
• An API for every tool
• OpenAI-compatible, with a key for each member.
• An admin panel
• Invite people, approve computers and label models.
• 21 harnesses
• For coding and security testing, each sealed in its own VM.
• Images, video and speech
• Media models take jobs through the same API.
• JSON on demand
• Send a schema and the server enforces it.
• Smart routing
• Each request goes to the least busy computer.
• Failover
• Up to three computers try a request before your app sees an error.
• Cached input
• A conversation stays where its context is already cached.
• A GPU dashboard
• Use, memory, temperature and power of every GPU.
• Uncensored labels
• Each model marked censored or uncensored.
• Prices and payments
• Make it public, charge per token and get paid into your wallet.

Your inference at work.

Requests, harnesses, media jobs and computers, all at your own address. A morning, for example:

• 09:02
• Request answered
• Served by the design workstation in 1.2 s
• 09:03
• Harness started
• In a sandbox VM that reaches only your inference
• 09:05
• Image job done
• 4 images, made in server room 1
• 09:06
• Computer offline
• Its requests moved to the studio PC
• 09:08
• JSON answer
• Checked against your schema by the server
• 09:09
• Computer approved
• Reception desktop, with its 8-character code
• 09:11
• Changes approved
• 2 files written into the folder, as reviewed
• 09:12
• Member joined
• With an invitation from your admin
• acme.yougori.com
• Live

Your apps switch in two lines.

Anything that speaks the OpenAI API works with your inference. Change its address and its key, and keep the rest.

• https://api.your-provider.com/v1
• YOUR_PROVIDER_KEY
• https://acme.yougori.com/v1
• YOUR_MEMBER_KEY
• Before
• After
• The same requests
• Chat, streaming, JSON and media requests work as they do now.
• A key per person
• Each member makes their own key, so usage is counted per person.
• Your models listed
• Ask /v1/models and your inference lists every model your computers share.
• Your app’s settings
• base_url
• api_key
• model
• stream
• "your-model"
• true

Questions.

What does it mean to create inference?

Creating inference means giving your AI models a place to answer requests: computers that run the models, an API your apps and tools call, and accounts and keys that decide who may use it. Yougori puts all of it behind one website at your own address.

Can you create our inference for us?

Yes. Tell us which models you want, who will use them and what hardware you have or want to rent. We set up the inference with you, connect your computers, tune the models and hand you the admin panel.

Where does our inference run?

On computers you choose: your own machines with NVIDIA GPUs, GPUs you rent from RunPod, Vast.ai, Lambda, Vultr or TensorDock, or a mix. Yougori hosts the website and the API address. The models run on your compute.

Can you help us get GPUs?

Yes. As your GPU broker we find the GPUs your models need, to buy or to rent, and connect them to your inference. Tell us your models, your users and your budget.

How much faster are your megakernels?

On average our megakernel writes 1.68 times as many tokens per second as others on the same GPU for one chat, 1.40 times with two chats at once and 1.35 times with three. Each megakernel is written for one model and one GPU family.

Do you guarantee the speed?

Yes. If our megakernel does not run your model at least 1.5 times as fast as your current setup on the same GPU, for one chat, you do not pay for the megakernel work. With several chats at once the gain is smaller, and we measure it with you before we start.

Which tools can use our inference?

Any harness or tool that speaks the OpenAI API. Each member creates their own API key on your inference website.

Can we set it up ourselves?

Yes. Create inference on yougori.com takes you through the company, access and compute steps, and the Yougori app connects each computer. Ask us when you want help or a megakernel for your model.

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 COMPANY. YOUR INFERENCE.

Tell us about your inference.