Cloud Horizon AI. Private beta opening summer 2026.

The LLM API that stays in Europe.

One API key, six open-source models, served from EU data centers under Dutch law. Drop-in replacement for the OpenAI SDK. Twelve dollars a month for personal use, fair-use rate limits, pay per token after that. No training on your prompts. No US parent.

Join the waitlist Read the docs

Waitlist members get first access and a year at half price. See what we shipped · Pricing · Security · Compare · Quickstart.

Target P50 latency: < 200 ms; first token, EU client
Models at launch: 6+; open weights only
Log retention: 0 to 7 days; you pick
Operated under: NL law; Spot Cloud B.V.

Drop-in for the OpenAI SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cloudhorizons.ai/v1",
    api_key="ch-...",
)

resp = client.chat.completions.create(
    model="kimi-k2.5",
    messages=[{"role": "user", "content": "Summarize this changelog"}],
)
print(resp.choices[0].message.content)

Same SDK, same call, EU residency. Swap one line.

Why we are building this

Three things every EU team asks before they put their prompts somewhere.

EU-sovereign by default

Inference and logging stay inside EU regions. Operated under Spot Cloud B.V. (KvK 89708873), a Dutch entity. No US parent, no foreign subpoena risk, no CLOUD Act exposure.

Zero training, zero retention

Your prompts and completions are not used to train any model. Logs auto-purge after 7 days, or 0 days on the team plan. Bring your own key for total isolation.

OpenAI-compatible API

Drop-in replacement for the OpenAI SDK. Swap the base URL, keep the code. Works with the official Python and Node clients, LangChain, LiteLLM, and the rest.

Built for the teams putting AI into production

Four buyer profiles we are designing for. The product is the same. The compliance story shifts with the audience.

DevOps and platform teams

Wire LLM tooling into runbooks, incident response, and infra automation without sending logs to a US gateway.

Auto-summarize PagerDuty incidents
Generate Terraform from intent
Review pull requests in CI

Read the deep dive →

Microsoft Certified Trainers

Build practice exams, draft course material, and grade lab submissions with a model that respects student data residency rules.

Generate AZ-104 practice questions
Translate course content to seven EU languages
Auto-score lab walkthroughs

Read the deep dive →

Regulated industries

Healthcare, public sector, finance. Auditable inference logs, DPA on day one, no data leaves the EU.

NL Schrems II compliant by design
BSI C5 control mapping available
EU AI Act inventory ready

Read the deep dive →

EU SaaS founders

Ship AI features to European customers without the awkward "your data goes to OpenAI" conversation in every sales call.

Customer support copilots
In-app document Q&A
Multilingual onboarding

Read the deep dive →

Model lineup at launch

Open weights only. We add the next strong release within 30 days. No hidden routing, no closed models behind the same name.

See the full model catalog with strengths, weaknesses, and the recommended use case per model.

Model	Context	Best for	Released
Kimi K2.5	200K	Long-doc reasoning, code review	Jan 2026
MiniMax M2.5	128K	Multilingual chat, agent loops	Feb 2026
GLM 4.6	128K	Tool use, structured output	Mar 2026
Qwen 3 Coder	64K	Pure code generation	Dec 2025
Llama 3.3 70B	128K	General-purpose chat	Dec 2024
Mistral Large 3	128K	European language fluency	Apr 2026

Drop in, same code

OpenAI-compatible endpoint. Pick the model on the request, swap the base URL, and you are done. Works with every official client and most community ones.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.cloudhorizons.ai/v1",
    api_key="ch-...",
)

resp = client.chat.completions.create(
    model="kimi-k2.5",
    messages=[{"role": "user", "content": "Summarize this changelog"}],
)
print(resp.choices[0].message.content)

import OpenAI from "openai"

const client = new OpenAI({
  baseURL: "https://api.cloudhorizons.ai/v1",
  apiKey: process.env.CLOUD_HORIZONS_KEY,
})

const resp = await client.chat.completions.create({
  model: "minimax-m2.5",
  messages: [{ role: "user", content: "Draft a release note" }],
})

console.log(resp.choices[0].message.content)

curl https://api.cloudhorizons.ai/v1/chat/completions \
  -H "Authorization: Bearer $CLOUD_HORIZONS_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-4.6",
    "messages": [
      {"role": "user", "content": "What is the EU AI Act, in two lines?"}
    ]
  }'

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://api.cloudhorizons.ai/v1",
    api_key="ch-...",
    model="qwen-3-coder",
)

result = llm.invoke("Write a Terraform module for an S3 bucket with KMS")
print(result.content)

Where your inference runs

Two EU regions at launch, two more by end of year. No US fallback, no third-country routing.

AMS

Amsterdam

Netherlands

Primary

FRA

Frankfurt

Germany

Primary

PAR

Paris

France

Q3 2026

STO

Stockholm

Sweden

Q4 2026

How we compare

Honest table. We are not the cheapest, we are not the most powerful. We are the one that keeps your prompts in Europe.

Capability	Cloud Horizon AI	OpenAI	Synthetic.new
Data stays in EU	Yes, by default	EU residency on Enterprise tier only	No, US-hosted
No training on your data	Yes, contractually	Opt-out, defaults vary	Yes
Model lineup	Open weights only	Closed GPT family	Mixed open weights
Personal plan	$12/mo + usage	$20/mo, no API	$12/mo + usage
OpenAI SDK compatible	Yes	Native	Yes
Operating entity	Spot Cloud B.V. (NL)	OpenAI L.L.C. (US)	Synthetic Inc. (US)
CLOUD Act exposure	No	Yes	Yes
Currency billed	EUR or USD	USD	USD

Pricing, plain

Two tiers at launch. Both come with EU residency, no training, and log retention you can set to zero.

Personal

$12 /mo

For solo builders

Personal API key
Fair-use rate limits, around six hours of heavy daily use
Pay per token beyond the limit, under OpenAI equivalent
7-day log retention
EUR or USD billing

Team

From $50 /mo

For squads and small companies

Shared keys, per-seat usage, SSO
Zero log retention option
Private endpoints and dedicated GPU hours
Data Processing Agreement, EU-only routing
Audit log export

Get on the waitlist

Three fields. We email you when the private beta opens, and we drop you a year at half price.

EU data residency from day one
OpenAI-compatible, swap the base URL
No training, no retention beyond what you set
Operated by Spot Cloud B.V. under Dutch law

FAQ

Why another LLM gateway? +

US gateways host EU prompts on US infrastructure under US legal reach. We run open-source weights on EU hardware, billed in EUR if you want, governed under Dutch law. That difference matters for healthcare, government, and regulated finance.

Which models do you serve? +

Kimi K2.5, MiniMax M2.5, GLM 4.6, Qwen 3 Coder, Llama 3.3 70B, and Mistral Large 3 at launch. We add the next strong open weights model within 30 days of release. No closed models, no per-token markup hidden behind a flat fee.

How is pricing structured? +

Twelve dollars a month gets you a personal API key with a fair-use rate limit. Team plans start at $79/month and add SSO and audit log streaming. Per-token rates apply on top, with the same rate on every plan. The dedicated /ai/pricing page has the full breakdown plus a token-economics calculator.

Can I bring my own GPU or model? +

Yes. The team plan supports private endpoints, including dedicated GPU hours and self-hosted weights. Talk to us if you need air-gapped deployment or a specific compliance posture.

How does this compare to Mistral La Plateforme? +

Mistral serves their own models well and lives in France. We run a wider open-weights catalog (Kimi, MiniMax, GLM, Qwen, Llama, Mistral) behind one OpenAI-compatible endpoint, so you can switch models without rewriting client code. Pricing is closer to per-token cost.

What is your data processing posture? +

Standard EU DPA available on team plans. Data stored only in EU regions (AMS and FRA at launch). Sub-processors disclosed. No US-based processing or storage at any point in the request lifecycle.

When does it launch? +

Private beta opens summer 2026. Waitlist members get first access and a year of the personal plan at half price. Public launch follows once we hit our latency and uptime targets.