ALLSmartLLM — The smart layer above the model

layer

Inference is the raw material. This is the layer that makes it usable.

The smart part was never the model. It's what surrounds it: routing, policy, privacy, proof.

One API, your aliases

An alias is more than a model name — it bundles model, provider and privacy policy. Swap the model centrally; your application never changes.

big-fat-model ──► kimi-k3 kimi-k3.1

// alias updated · zero code changes · apps keep running

Privacy in stages

Mask, pseudonymize, or go strict: an LLM on our own hardware double-checks for indirect identifiers before anything leaves the EU.

mask pseudonymize strict

Please reply to <<PERSON_1>> from <<ORG_1>> regarding invoice <<IBAN_1>>

Mapping lives in request memory only. Never stored, never logged.

Your data stays in Europe

Our own hardware in Austria. No sub-processors, no third countries. External frontier models are reachable — but only through the masking layer.

your app ──► SmartLLM · AT ─▍ third country

Prompts terminate on hardware we own — or leave masked, by your policy.

Measured, not promised

Nightly benchmarks with published methodology. Detection recall for the privacy layer as a number, not an adjective.

Qwen 3.5 397B

26.5 t/s

Nemotron 3 Omni

39.9 t/s

Nemotron Cascade 2

40.2 t/s

Decode rate, streaming, median of repeated nightly runs.

start

First response in under five minutes

If your software speaks OpenAI, it already speaks SmartLLM. Change one URL.

step 1

Create your account

10M tokens free to test and set up. No card, no subscription.

free tokens10,000,000

step 2

Get your API key

One key, all models, all aliases. Usage-based, nothing else.

SMARTLLM_API_KEY••••••••••••

step 3

Make your first request

Call an alias, not a model version. Privacy is one parameter.

curl https://api.smartllm.at/v1/chat/completions \
  -H "Authorization: Bearer $KEY" \
  -d '{
    "model": "frontier",
    "privacy": {"mode": "pseudonymize"},
    "messages": [...]
  }'

Later, host it yourself: same API on your own hardware. When your box is running, you change one URL — done.

models

Curated open-weights models

Vetted, benchmarked, and kept current. Aliases follow our recommendation — or pin a version and stay.

Model	Alias	Context	Measured	Status
Qwen 3.5 397B	frontier	256K	26.5 t/s	loaded
Nemotron 3 Omni	—	128K	39.9 t/s	on-demand
Nemotron Cascade 2	—	256K	40.2 t/s	on-demand
Phi 4	—	16K	—	on-demand
Qwen 3.6 35B	mid vision coder	32K	53.5 t/s	loaded

15 models · all on dedicated hardware in Upper Austria View all models & nightly benchmarks →