How it works
Most gateways speak the OpenAI API format. To use one, you change two settings in your code: the address and the key. The rest of the code stays as it is.
# the same OpenAI SDK, pointed at a gateway
from openai import OpenAI
client = OpenAI(base_url="https://gateway.example/v1", api_key="GATEWAY_KEY")
reply = client.chat.completions.create(
model="provider/model-name",
messages=[{"role": "user", "content": "Summarise this log"}],
)For every request, the gateway does four things:
- Checks the key and how much it may still spend.
- Picks the provider for the model you named, and tries another one if the first is down.
- Passes the request on and streams the answer back.
- Writes down the tokens, the cost and the time.
A gateway or a direct key
| A key from each provider | An AI gateway | |
|---|---|---|
| Keys to keep | One per provider | One |
| Code per provider | Each has its own format | One format for all |
| A provider goes down | Your app fails | The request goes elsewhere |
| A budget per user or app | You build it | A key with a cap |
| Bills | One per provider | One |
| Who sees your prompts | The provider | The provider and the gateway |
Kinds of AI gateways
| Kind | Example | What you get |
|---|---|---|
| A model marketplace | OpenRouter | 500+ models with one key and one balance, and a 5.5% fee when you buy credits on the pay-as-you-go plan |
| A layer over your own keys | Cloudflare AI Gateway | It sits in front of your own provider keys and adds caching, spend limits, rate limiting, routing and logs |
| Software you run yourself | LiteLLM | It is open source, you host it yourself, and it calls 100+ models in the OpenAI format |
| Built into your computer plan | Uno AI | A fast and a smart model paid in hours of work, other models per token, and the computer your agents run on |
Do you need one?
Take a gateway when
- You use more than one model. A cheap one for routine calls and a strong one for hard ones.
- You give AI to other people. Each user or app gets a key with its own spending cap.
- Agents run unattended. When a provider fails, the gateway retries elsewhere instead of stopping the job.
Skip it when
- One app uses one model. A direct key is simpler.
- You only use Claude Code or Codex yourself. Your Claude or ChatGPT plan already covers them.
- Your data must go only to the model provider. A gateway is one more company that sees your prompts.
An AI gateway in Uno
Every Uno account has one at https://console.uno.place/v1. It is OpenAI-compatible: point any OpenAI SDK at it. Fast and Smart are paid in Uno AI hours; Claude, GPT and Gemini per token. For your own users there are child keys: a separate key for each user or app, each with its own spending cap; past the cap the key answers 402.
Uno AI: +$15 a month on Small
A gateway your agents use without counting tokens
Uno AI writes code, builds sites and runs tasks right on your computer. Only the minutes it works count; ten chats in the same minute count as one.
Today: GLM-5.3 Flash (Fast) and MiMo v2.6 Pro (Smart). Dictation into text at no extra cost, within a daily limit.
- 40 hSmall + AIa month
- 100 hPlus + AIa month
- 45 h + FastPro + AIfull speed, then Fast never stops
- UnlimitedMax + AI100 h at full speed, then standard
Pick a computer, then add Uno AI
Every plan is a real Linux computer with root. Your agents run on it and call models through one key.
Small
A Telegram bot, a VPN, a small backend
$5/ month
- 1 vCPU · 1 GB
- 15 GB disk
- 25 GB cloud storage
- 5 boosts a month
Plus
Agents that work with your laptop closed
$20/ month
- 4 vCPU · 4 GB
- 40 GB disk
- 100 GB cloud storage
- Uno Work in the cloud
Pro
A product in production, a second MacBook
$70/ month
- 16 vCPU · 16 GB
- 120 GB disk
- 300 GB cloud storage
- Speed of a MacBook Air M4
Max
A team and many agents at once
$200/ month
- 32 vCPU · 48 GB
- 300 GB disk
- 1,024 GB cloud storage
- 5 public IPv4
Add Uno AI to any plan: 40 h of AI work a month on Small, 100 h on Plus, unlimited on Max. Today: GLM-5.3 Flash and MiMo v2.6 Pro. Or sign in with your own Claude or ChatGPT.