NullRouter
Get API access

Inference thatleaves no trace.

Multi-cluster LLM inference built in Europe. Prompts exist only in memory — never on disk.

Get API accessHow it works

api.nullrouter.dev/v1 · OpenAI-compatible

0

bytes written to disk

0

independent GPU clusters

0

OpenAI-compatible endpoint

0%

EU-hosted request path

Platform

Nothing can leak. By design.

Four guarantees enforced by the architecture — not by policy.

OpenAI-compatible

Keep your stack.

Point the official SDKs at the API. Only the base URL and key change.

base_url = "…/v1"

api_key = "nr_live_••••"

Blind relay

Ciphertext only.

The EU relay forwards encrypted bytes. It holds no certificate and no keys.

a3 7f e0 12 9c 4b d1 88 fe

no keys held at the relay

Multi-cluster

Failover built in.

Two independent clusters behind one endpoint. Health checks reroute automatically.

cluster-1ready
cluster-2ready
Per-key budgets

Caps, not surprises.

Allowlists, rate limits and monthly budgets in EUR — enforced at the gateway.

monthly budget62%

enforced at the gateway

Request path

Decrypted in exactly one place.

Five hops. Everything upstream of the gateway sees only ciphertext.

  1. 01

    Client

    OpenAI SDK

    TLS 1.3

  2. 02

    Relay · EU

    EU region

    ciphertext only · no keys

  3. 03

    Tunnel

    private mesh

    private route

  4. 04

    Gateway

    our hardware

    TLS terminates · auth · routing

  5. 05

    Cluster

    vLLM

    streams tokens

The relay and the tunnel see encrypted bytes. The gateway holds your prompt in RAM for the life of the request — then it's gone.

Models

Stable names. Independent clusters.

Two open models today — served from separate GPU clusters behind one endpoint.

deepseek-v4-flash

cluster 1

General-purpose speed for reasoning, writing and code.

  • text only
  • streaming tokens
  • cache isolated per key

deepseek-v4.1-flash

cluster 2

Latest generation, with vision.

  • text + vision
  • one inline base64 image
  • remote URLs never fetched

Upgrades happen behind the API — your integrations never change.

Security · zero retention

Zero retention is a release gate, not a feature flag.

A canary string is pushed through the live path, then hunted across every disk, journal, container and database. Expected result: zero matches.

journald

0

containers

0

postgres

0

wal

0

/tmp

0

canary → clean

logs offcache offcallbacks offtraining off

Pricing

Metered in tokens. Billed in euros.

Every key carries a monthly budget in EUR — enforced at the gateway before a request reaches a GPU.

Input tokens20M
5M – 500M€0.25 / 1M*
Output tokens2M
0.5M – 50M€1.00 / 1M*

Estimated / month

€ 7.00*

€0.25 / 1M input · €1.00 / 1M output*

Billed monthly · input + output combined.

Get API access

* Starting prices. Final rates depend on the model and may change with demand.

FAQ

Straight answers.

Prompts are decrypted at the gateway and held in RAM while the request runs. Only metadata — token counts, latency, status — is stored.

Get started

Ready when you are.

Tell us what you're building. We'll set up a key with the right models, limits and a monthly budget — and answer your security questions first.

Email hello@nullrouter.dev

api.nullrouter.dev/v1 · OpenAI-compatible · DPA available