One API for every model.
Drop-in and OpenAI-compatible: swap your base_url and you are behind the lock.
Sluis is the gateway in between: one API for every AI model that strips personal data from each prompt before it leaves, puts it back in the answer, and seals every passage in an audit ledger.
First 50,000 tokens free
Now supporting Grok by xAI. One line of code. +1 ms measured overhead. No seats, no minimums.
Drop-in and OpenAI-compatible: swap your base_url and you are behind the lock.
Detection and reversible pseudonymization before anything leaves; the key never exits your perimeter.
A sealed, hash-chained ledger records every passage, ready for your auditor.
Call sluis/auto or a use-case alias; Sluis refreshes managed routes daily from Artificial Analysis and resolves the target inside your residency policy.
A lock does what no dam can: it holds back the sea and lets the ship through. It never opens both gates at once, and every passage goes in the keeper's ledger. Sluis does the same for AI traffic: your data sails through, what must stay inside stays inside, and every passage is provable.
Sluis is Dutch for canal lock.
The same request: as your user types it, as the model receives it, and as your app gets it back.
No request skips a gate. None leaves without a record.
60+ detectors scan every prompt before it leaves; what they find is reversibly pseudonymized.
You decide per key: which models, which region, whose ownership; a failed provider fails over automatically.
Every request gets a hash-chained entry: verifiable offline, exportable, erasable per person (GDPR Art. 17, or any right-to-delete duty).
An address says where a server stands, not who owns it. Legal reach follows ownership: a provider's parent company can be compelled under laws like the CLOUD Act.
Where a model runs and who owns it are two different risks; Sluis labels both axes on every model in the catalog.
The default is the strictest setting: EU-owned providers on EU-region clouds only, until you deliberately allow more.
One key allows a frontier model for general drafting; another stays locked to the strictest route for anything personal.
If your models or agents serve EU customers, Sluis is your compliance bridge: put the lock in front of your product and every call gains enforced EU residency, reversible pseudonymization, and a verifiable audit trail. No rebuild, one base_url.
Whether you ship the code or sign off the risk, Sluis is built for you.
Swap one base_url and you are behind the lock: OpenAI-compatible, streaming stays streaming, +1 ms measured.
Read the docsDPA, sub-processor list, DPIA input, and answers to your security questionnaire.
MCP tool calls pass the same three gates as any request: an agent reading your CRM is as watertight as your chatbot.
tool.crm.readLIVEgrant.customer.readALLOWanna@bedrijf.nl → «EMAIL_1»MASKroute.eu-ownedEUsha256:8f21…b40aSEALEDThe whole data plane as a single binary on your own iron: same gates, same ledger, your perimeter.
Switch from Chat to Passage. Sluis turns the goal into a Passage Contract, routes it to an owner-activated Station, keeps work inside a Chamber, pauses sensitive actions at a Gate, and gives every successful verified outcome a Seal.
passage-contract.v3Agree the goal, limits, acceptance checks and budget before work starts.
An owner explicitly activates a macOS, Windows or Linux computer for execution.
Folders, applications, tools and collaborators stay inside one bounded workspace.
Shell, browser, computer and other high-impact actions pause for policy or human review.
Every outcome carries the exact contract, actions, approvals, changes and checks.
Your provider's list price, plus a flat 10%. Your first 50,000 tokens are free.
See pricingCompliance shouldn't mean giving up the best models. That's a false dilemma. You're just missing a lock.
Use Chat for an answer or Passage for durable work. A Station executes inside a Chamber, risky actions stop at Gates, reusable Skills standardise the workflow, and every successful verified Passage ends with a Seal.
Open Sluis Chat →