Onboarding the first hosts and builders

Someone’s idle GPU.
Your unlimited AI.

Millions of idle GPUs and CPUs in homes and offices, pooled into one network that runs AI.

Waiting for request
0 tokens—

Host & earn

Turn idle hardware into income.

Run a model on your gaming rig, Mac mini, Mac Studio or datacenter GPU. Join the network, process requests and get paid per token.

Benchmarked on join
Your machine is tested and ranked the moment it's online.
Paid per token
Every token your node generates counts toward your earnings.
A gaming PC with RGB lighting
Node previewServing

Gaming rig

8 GB VRAM

18,342

tokens served

GPU memory

e.g. RTX 3060 Ti, RTX 4060
Apple silicon? Count unified memory.

  • Qwen 2.5 Coder 32B

    24 GB RAM · 24 GB VRAM

    Needs 24 GB

  • Gemma 2 9B IT

    12 GB RAM · 8 GB VRAM

    Ready

  • Llama 3.1 8B Instruct

    12 GB RAM · 8 GB VRAM

    Ready

  • Nous Hermes 2 Mistral 7B DPO

    10 GB RAM · 6 GB VRAM

    Ready

  • Stable Diffusion XL

    16 GB RAM · 8 GB VRAM

    Soon

3 of 4 live models ready on this machine.

1import os
2from openai import OpenAI
3
4client = OpenAI(
5−    base_url="https://api.openai.com/v1",
6+    base_url="https://api.ukoo.network/v1",
7    api_key=os.environ["UKOO_API_KEY"],
8)
9
10reply = client.chat.completions.create(
11    model="llama-3.1-8b-instruct",
12    messages=[{"role": "user", "content": "Habari!"}],
13)

Credits

4,812,330tokens left

Prepaid
No usage caps

Build without caps

Prepaid, uncapped AI.

Load credits once and use them without surprise bills. OpenAI-compatible endpoints mean any app, CLI or agent plugs straight in.

  • Swap one linePoint your OpenAI SDK at Ukoo. Keep the rest of your code.
  • Chat, code, agentsBuild chat interfaces, coding tools or niche agents on top.

Models

Open models, on real machines.

Each one is verified, benchmarked and routed to nodes that can actually run it.

All models

Qwen 2.5 Coder 32B

Coding

Active

24 GB RAM · 24 GB VRAM

Gemma 2 9B IT

General knowledge

Active

12 GB RAM · 8 GB VRAM

Llama 3.1 8B Instruct

Agents

Active

12 GB RAM · 8 GB VRAM

Nous Hermes 2 Mistral 7B DPO

Agents

Active

10 GB RAM · 6 GB VRAM

Stable Diffusion XL

Image

Coming soon

16 GB RAM · 8 GB VRAM

Agents leaderboard. Once the network is live, partner agents are ranked by real daily usage.

See agents

Pricing

One meter. Tokens.

No plans to decode. Builders pay per token and hosts earn per token.

If you build

Prepay credits.
Spend per token.

  • Load credits once, then build
  • Every request metered in tokens
  • No surprise bills and no caps

If you host

Serve requests.
Earn per token.

  • Benchmarked and ranked automatically
  • Routed requests the moment you're online
  • Paid for every token you serve

Join the ukoo.

ukoo (n.) · Swahili for clan, the wider family you belong to

Bring a machine or bring an idea. We’re onboarding the first hosts and builders now.

Join our Discord