SOKKAN Inference — sovereign EU inference for coding

Three speeds, one endpoint

You ask for a tier; we pick the best sovereign open-weight model behind it. Ship handles the everyday; Deep takes over on hard tasks — automatically.

Default

Ship

0.40 / 1.50

CHF / M tokens (in / out)

The coding workhorse: edits, completion, tool calls. As good as a frontier model in our tests, ~7× faster. model: sokkan-ship

Boost

Deep

4.00 / 20.00

CHF / M tokens

Frontier-open reasoning. The cavalry when Ship stalls — and the tier we escalate to on our own (unrequested escalation: billed at your tier's price, first 1M tokens/month). model: sokkan-deep

Switzerland

Swiss

0.60 / 2.40

CHF / M tokens

A 20B-class model, served in Geneva on our own hardware — your data never leaves Switzerland. When Geneva is saturated or down, requests overflow to a Swiss-hosted cloud (Infomaniak) — never outside Switzerland, never to an EU provider. Best for chat, RAG, agents; not frontier coding. model: sokkan-swiss

Need true frontier (Claude) on a specific project? Optional Frontier tier (BYOK, your key) — the offer stays 100% EU by default, Claude is a documented choice.

Fair billing, verifiable. Prepaid, no daily cap — your balance is the only limit. Exact per-token metering (no per-request minimum; centime fractions accrue to your credit). Balance, monthly spend and the escalation gauge are visible at any time in your cockpit and via GET /usage with your token.

What sets it apart

◈Sovereign by default

Inference served from EU datacenters, outside the US CLOUD Act. Zero retention, zero training on your data, DPA (GDPR art. 28). Your prompts only ever serve to answer you.

◈Never “not capable” — never at your expense

If your tier's upstream falters, we fail over to a second EU provider at the same price, then escalate to Deep as a last resort. Escalation is billed at YOUR tier's price for the first 1M escalated tokens each month — a gauge in your cockpit makes it visible. Escalating earns us nothing: it's our quality insurance, not your bill.

◈Secrets blocked, zero latency

A deterministic guard scans every prompt before it’s sent (~a few ms, no AI call). An API key, a token, a private key? Rejected — nothing leaves. Personal data is masked.

Connect Claude Code in 30 seconds

It’s an Anthropic-compatible API. Point your tool at SOKKAN, pick a tier — that’s it.

# your SOKKAN inference token (prepaid, in CHF)
export ANTHROPIC_BASE_URL=https://infer.sokkan.ch
export ANTHROPIC_AUTH_TOKEN=sik_your_token
export ANTHROPIC_MODEL=sokkan-ship   # or sokkan-deep for the boost

claude "refactor this module and add the tests"

Already on the SOKKAN Cloud cockpit? Managed inference is built in — nothing to configure, balance and per-user usage in your dashboard.

Sovereign inference, for coding.