razorBridge AI Gateway
AI models by the token. You choose where they run.
One OpenAI-compatible API, with a price per million tokens and a residency label on every model. EU-only models run in Paris on an EU-headquartered provider and never fall back outside the EU.
From €0,15 per 1M input tokens · Mistral Small 3.2, EU-only
Products
Three products, one company behind them.
Start with an API key today. Bring kOS in-house, or rent a GPU, when you need one.
One API. Residency on every model.
OpenAI-compatible. EU-only models fail closed; every other model is labelled. €0,15 / 1M tokensFrom, input · Mistral Small 3.2, EU-onlykOS, hosted and run by us.
KlusAI's AI workspace in two tiers: Shared, or Dedicated with private inference. razorBridge OutpostDesign partnersOn site: kOS inside your network, over an outbound-only link. Per organisationPriced with you, invoiced by contractYour own GPU, SSH-ready.
NVIDIA L4, L40S and H100 by the hour. See the price, launch, SSH in. €1,09/hFrom, excl. VAT · 60-min minimumAI Gateway
Change the base URL. Keep your code.
Models are grouped by where they run. EU-only models run in Paris with an EU-headquartered provider and fail closed, with no fallback outside the EU. Every other model says so and names its provider.
- Input and output price per million tokens on every model.
- Self-serve keys draw on the same prepaid balance as your AI Blades.
# python -m pip install openai
import os
from openai import OpenAI
client = OpenAI(
base_url="https://razorbridge.eu/api/v1/compute/gate",
api_key=os.environ["RAZORBRIDGE_KEY"],
max_retries=0,
)
response = client.chat.completions.create(
model="razorbridge-eu/deepseek-v4-flash-0731",
messages=[{"role": "user", "content": "Hello!"}],
max_tokens=128,
)
print(response.choices[0].message.content)curl --fail-with-body https://razorbridge.eu/api/v1/compute/gate/chat/completions \
-H "Authorization: Bearer $RAZORBRIDGE_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"razorbridge-eu/deepseek-v4-flash-0731","messages":[{"role":"user","content":"Hello!"}],"max_tokens":128}'EU-only · fails closed
May be processed outside the EU
Per 1M tokens, excl. VAT, from the public model catalogue.
Managed kOS
kOS, hosted and run by us.
kOS is KlusAI's AI workspace for documents, knowledge and agents. Pick how much of it is yours alone.
- Priced per organisation and invoiced under a contract.
- Dedicated names every exception to EU hosting in your contract.
Shared
Multi-tenant and EU-hosted. AI requests may be processed outside the EU.AI requests may leave the EU
Dedicated
Single-tenant, EU-hosted with named exceptions. Includes private inference: GPUs that serve only your organisation.Exceptions named in your contract
Priced per organisation.
Managed kOS, on site
razorBridge Outpost. kOS, inside your network.
A design-partner programme for connecting on-premises systems to razorBridge Managed kOS. Encrypted outbound connectivity and on-site AI are planned capabilities, to be validated with partners.
Design render
AI Blades
Pick a GPU. See the price. SSH in.
From €1,09/h excl. VAT, 60-min minimum
Usually ready in a few minutes. Auto-stop means a forgotten GPU stops billing, and the Hub tells you how long your balance lasts before you launch.
- Billed per second after a 60-minute minimum, excl. VAT.
- A plain SSH host: scp, rsync and VS Code Remote work.
- Multi-GPU classes are on the pricing page.
Unified-memory Macs for private inference.
L4
24 GB VRAM€1,09/h
first 60 minutes: €1,09 minimum
Light workloads & demos
L40S
48 GB VRAM€1,99/h
first 60 minutes: €1,99 minimum
Training & workshops
H100
80 GB VRAM€4,29/h
first 60 minutes: €4,29 minimum
Large model training
Drawn from the Hub's launch view. Prices are live; the layout is illustrative.
The Hub
This is the product, not a mock-up.
The razorBridge Hub is where you launch AI Blades, plan labs and events, and create API keys. Every GPU shows its price and its billing minimum before you start it.
A screenshot of the razorBridge Hub with demo data.Built and operated by KlusAI Labs SRL, Cluj-Napoca, Romania.
Built for
Built for the way you buy compute.
Five kinds of buyer, five different first steps.
Price, speed and an API you already know.
- An OpenAI-compatible endpoint: swap the base URL.
- Prices on the page, before you start.
- SSH, scp, VS Code Remote: a GPU is a plain host.
A GPU lab for your class, paid by invoice.
- Students join with a link or QR code. They never enter a card.
- Pay by pro-forma or invoice from KlusAI Labs SRL.
- The lab tears itself down at the end time.
One GPU per team, gone when the event ends.
- Estimate the total before you book.
- Teams join with a code at the door.
- A live board: who joined, which GPU, spend so far.
GPU 1L4
GPU 2L4
GPU 3L4
GPU 4L4
GPU 5L4
GPU 6L4
Illustrative live board
Residency you can read, line by line.
- Managed kOS Dedicated: single tenant, private inference.
- Every model labelled EU-only or not.
- Sub-processors named in the DPA.
IllustrativeYour systems stay where they are.
- Planned: outbound-only connectivity
- Planned: on-site inference and embeddings
- Design-partner qualification is open.
Design renderEvents and labs
A GPU lab for your class, in one link.
For courses, workshops and hackathons, built on AI Blades. You plan it, students join, and it cleans up after itself.
GPU 1L4
GPU 2L4
GPU 3L4
GPU 4L4
GPU 5L4
GPU 6L4
Illustrative live board
01 Plan
See the total first.GPU, seats and hours. The estimate is the event's cost before VAT.
02 Share
A link and a QR code.Students sign in and join. Nobody enters a card.
03 Done
It tears itself down.GPUs stop at the end time.
What will it cost?
Residency and trust
Where your data is processed, product by product.
No blanket promises. Each product says what stays in the EU and what may not.
Illustrative. razorBridge runs in data centres operated by its cloud providers.
EU-only models run in Paris with an EU-headquartered provider and fail closed. Every other model is labelled, with its provider named.
Labelled per model
EU-hosted. AI requests may be processed outside the EU.
AI requests may leave
Single-tenant, EU-hosted with named exceptions listed in your contract. Includes private inference.
Named exceptions
Planned for your premises, through the Outpost design-partner programme. Connectivity and local AI remain to be validated; synced records would be stored in kOS.
On your premises
EU data centres only, operated by EU-headquartered cloud providers.
Stays in the EU
Pricing
Prices you can see before you start.
All prices exclude VAT. Prepaid balance for self-serve; invoice for universities and contracts.
€0,15
Per organisation
€1,09/h
€1,99/h
€4,29/h
€19,62
Questions
Straight answers.
Something else? Talk to us.
Which AI models stay in the EU?
The ones labelled EU-only. They run in Paris with an EU-headquartered provider and fail closed rather than fall back outside the EU.
How does billing work?
Self-serve accounts use a prepaid balance. Each GPU bills at least 60 minutes, then per second, and AI models bill per million tokens. Universities and contracts pay by invoice. You see the price before every start.
What is kOS, and do I need it?
kOS is KlusAI's AI workspace for documents, knowledge and agents. You don't need it for AI Blades or the AI Gateway.
Is Outpost available?
To design partners. Tell us about your setup.
Where do the GPUs run?
In EU data centres only, operated by EU-headquartered cloud providers.
Can a university pay by invoice?
Yes. We issue a pro-forma and an invoice from KlusAI Labs SRL, with reverse charge for EU VAT-registered institutions.
Do students need an account or a card?
They need an account to join, but never a card. The event's organiser pays for its GPUs.
What happens when a lab or event ends?
Its GPUs stop at the end time, and billing stops with them.
Get started
Change the base URL. Keep your code.
Create a key and send your first request.