För AI-native team · EU-mark

Host your LLM apps in Sweden.

Your inference API, vector DB, agents and RAG pipeline — running on Swedish soil, under EU law, on hardware we own. Build on compute you control — with GPU inference on 6× L40S, live in Falkenberg today.

Per-tenant isolationEU/Sweden data residencyAPI + Terraform
API layer (your app)gp1.large · ubuntu-24.04
Vector DB + storagepgvector · NVMe block
GPU inference6× L40S · live — book a card
Why Kepler for AI

Your model weights are a trade secret. Keep them home.

EU data sovereignty

Weights, embeddings and prompts stay in Falkenberg — under Swedish and EU law. No US Cloud Act.

Per-tenant isolation

Your own OpenStack project and credentials, rotated by OpenBao. A checkable security proof.

API + Terraform first

Stand up your whole inference stack as code. Scale vertically without leaving your Planet.

Ship in three steps

From repo to running inference API.

1

Spin up compute

Create a gp1.large Planet with a curated image. SSH key injected at boot.

2

Deploy your stack

vLLM/TGI + pgvector behind your API. Private VPC between app and DB.

3

Go live on EU soil

Floating IP, security groups, snapshots. Book an L40S and run inference on EU soil.

~/llm-app · deploy
$ kepler instance create –flavor gp1.large –image ubuntu-24.04
→ i-7a3f active (47s) · floating_ip 194.x.x.18
$ docker compose up -d vllm pgvector api
→ inference API live · https://api.yourapp.eu
Straight talk: 6× L40S are live and single-tenant — the whole card is yours, never shared. Bigger capacity (H100-class) is on the roadmap, and we only say ”now” about hardware that’s actually in the rack. See the roadmap →

Book an L40S — live in Falkenberg today.

Run inference on your own card, on EU soil, starting this week. Need H100-class? Join the list and we’ll reach out when bigger capacity lands.

No spam. An engineer replies within one business day — H100-class interest goes on the list.