AI Routers
One gateway for all your AI traffic. Send every request to a single endpoint, define rules and classifiers to pick the right model each time, and fall back automatically when a model fails, with no infrastructure to build or maintain. Stop paying premium prices for simple prompts and keep your AI costs under control by using your most powerful models only when it really matters.
- Status
- Coming soon
- Category
- AI & Data
- Access
- Early access
- Release date
Key facts
- For
- developers and teams running AI applications in production
- Alpha
- December 2026
- Price
- no extra charge during Alpha, you pay only for the AI Endpoints tokens your requests consume
Build your flow in a visual editor, or by API
A router is a flow of steps: a Switch or an If branches on what a classifier or a built-in signal says about the request, and every branch ends with a result, a Model and its fallbacks, or an Error you define. Build it in a few clicks in the OVHcloud Console, or with one call to the OVHcloud API.
The cost-saver router, with one Switch on the built-in Complexity signal:
Why a router?
Most apps send every request to the same model, often the largest one, just in case. Yet a short translation or a classification does not need a 70-billion-parameter model: a small one answers as well, faster, for a fraction of the price. And a request with code is better served by a coding model.
AI Routers reads each request and picks the model that fits it, with your rules. Your apps do not change: they call one OpenAI-compatible endpoint, with the router name in the model field.
from openai import OpenAI
client = OpenAI(base_url="https://ai.router.cloud.ovh.net/v1", api_key=ROUTER_KEY)
client.chat.completions.create(
model="cost-saver", # your router, not a model
messages=[{"role": "user", "content": "Translate 'see you tomorrow' into Spanish"}],
)
Everything around the model, managed
One endpoint
OpenAI Chat Completions and Responses, Anthropic Messages, streaming included. Your SDKs and coding tools work as they are.
Rules and classifiers
Built-in signals (complexity, code, prompt length) answer in under a millisecond. Add your own keywords, regular expressions or an AI classifier.
Automatic fallbacks
Each model has its fallbacks. Retries and circuit breakers skip a failing or saturated model without your app noticing.
Cost control
Keep your most powerful models for the requests that need them, and send the rest to smaller, cheaper ones.
Test before you ship
Explain a prompt: see the path it takes and the model it would reach, without calling the model. Free, and nothing is stored.
Activity
Every decision, logged: which path, which model, how long.
Templates
Start from a ready-made router, such as Cost saver, or from scratch.
Sovereign
The router and the AI Endpoints models run on OVHcloud, hosted in Europe.
Questions, answered.
In December 2026.
OVHcloud AI Routers will be available from the OVHcloud Console in December 2026, under Public Cloud > AI > AI Routers, and through the OVHcloud API.
AI Routers opens as an alpha in December 2026. It will be under active development: the interface and APIs may change before beta and General Availability, and there is no SLA during the alpha.
AI Routers adds no extra charge during the alpha. You are billed only for the AI Endpoints tokens your requests consume, according to the current AI Endpoints pricing. Testing a prompt with Explain is free: the model is not called.
During the alpha, the models of the OVHcloud AI Endpoints catalogue, called with your AI Endpoints API key.
No. AI Routers exposes one endpoint compatible with OpenAI Chat Completions and Responses and with Anthropic Messages, streaming included. Point your SDK or tool at it and put your router's name in the model field.
Its fallbacks answer, in the order you set. Retries and circuit breakers skip an overloaded model without your application noticing.
No. As with our other AI products, your data is never used to train or improve OVHcloud's AI models. Prompts tested with Explain are not stored.
Shape what ships next.
Join the User Experience Program to test upcoming products, get early alpha and beta access, and put your fingerprints on the roadmap.