01 Classified
Inspect
60+ detectors scan every prompt before it leaves. PII, PHI and secrets are reversibly pseudonymised before a single token moves on.
A SevenLab productSovereign AI gateway
Any model. Provably compliant.
Nothing personal ever leaks.
Sluis is a drop-in OpenAI endpoint. Change one line, and it strips every name and secret before the model sees them, puts them back in the response, and seals every call in a ledger you can't fake. GDPR compliant, EU-sovereign by default.
Your first 250K tokens are free. Nothing leaves your residency policy.
- Drop-in, change one base_url
- GDPR compliant, EU-sovereign by default
- Tamper-evident audit on every call
- +1 ms measured overhead, every gate on
Your app · request
Jane Doe · [email protected]
Sluis gate
«NAME_1» «EMAIL_1»
- 01 inspect · dlp
- 02 route · residency
- 03 seal · ledger
Route · any model, your policythe model only ever sees the token
EUmistral-large · eu-parisin-region
USgpt-5.1 · us-eastallowed · policy
CNdeepseek-v3 · cnblocked
Built for teams who answer to regulators
- Banks
- Health insurers
- Law firms
- Public agencies
- Fintechs
How it works
Every request clears three gates
A lock holds back the sea while it moves the boat. Sluis works the same way: nothing skips a gate, and nothing leaves without a record.
02 Routed
Route by policy
Your policy picks the provider. An EU-sovereign model first, an EU-region cloud as backup, a hard stop if nothing fits.
03 Sealed
Seal in the ledger
Every request gets a hash-chained entry: verifiable offline, exportable, erasable per data subject (GDPR Art. 17). Break one link and the whole chain shows it.
Reversible pseudonymisation
Work with real personal data. The model never sees it.
Sluis swaps every name, ID and secret for a stable token before the prompt ever leaves your network. Suddenly any model is safe to call, direct US and Chinese providers included, on opt-in. The real values come back in the response, and the token map is never stored.
- Names, emails, IDs and secrets tokenised on the way out
- Real values restored in the response, never kept around
- Any model becomes safe to call, even the ones you couldn't touch before
Your prompt
Refund customer Jan de Vries, IBAN NL91ABNA0417164300.What the model sees
Refund customer [PERSON_1], IBAN [IBAN_1].Residency policy
You decide where data can go. Sluis enforces it on every call.
Map your data classes to the regions and providers you trust. Sluis checks it on every request. No exceptions, no data quietly slipping out.
- EU-sovereign by default, US and China available on opt-in through explicit policy
- Fallback chains you define, logged separately so you always know when the backup ran
- Hard stops with a clear reason code, not a best guess
residency.policyclass: PHIenforced
- primarymistral-fr, mistral-large-latestEU, France
- fallbackscaleway-fr, llama-3.3-70bEU, France
- allowedvertex-eu, gemini-3.5-proEU region
- blockedopenai-us, gpt-5.1US, East
Tamper-evident ledger
The kind of record an auditor actually trusts, because it proves its own integrity.
Every entry is hashed with the one before it. Change a single field anywhere in the history and every link after it breaks. Export the chain and anyone can re-check it offline.
audit.chainlast 4 entries chain verified
- #4f9c2a← e1b7d9mistral-frPHIin-region
- #e1b7d9← 7a3f10scaleway-frPHIfallback
- #7a3f10← 0c55abmistral-frPIIin-region
- #0c55ab← genesismistral-frgeneralin-region
Model coverage
Sovereign by default. Any model when you need it.
Sluis shows you what other gateways hide: who owns each model and where it runs. EU-owned first. Frontier models like Claude and Gemini run today in EU regions. Direct US and Chinese providers are available too, on opt-in, each flagged for CLOUD Act exposure and only ever used when your policy says so.
Mistral
sovereignScaleway
sovereignGoogle Vertex
EU region, US-ownedAWS Bedrock
EU region, US-ownedOpenAI
opt-inAnthropic
opt-inDeepSeek
opt-inYour own model
bring your own+1 msmedian gate overhead, measured
Performance and reliability
The fastest gateway that can still prove where every prompt went.
A stateless hot path, with an in-process cache and a streaming tee that never buffers your response. Turn every gate on and Sluis adds +1 ms at the median in our published benchmark.
- One millisecond at the median, every gate on and measured
- Key pooling and failover that never break your residency policy
- Prometheus and OTLP metrics tagged with jurisdiction, owner and decision
Pricing
Your provider's list price, plus a flat 10%.
The fee is for the lock, not the tokens. No seats, no minimums, no markup games. Your first 250K tokens are free, and nothing leaves your residency policy.
See pricingCompliance shouldn't mean giving up the best models. That's a false dilemma. You're just missing a lock.
FAQ
Questions, answered
Swap one line today. Pass your next audit in your sleep.
Your first 250K tokens are free. Nothing leaves your residency policy.
Read the quickstart