What is Salus · Salus docs

The data protection layer between your organisation and external AI. It detects sensitive values inside your environment, replaces them with reversible tokens before a request leaves, and restores them when the answer returns.

Employees, applications and agents are already sending customer data to external AI services. The two conventional answers both fail. Blocking stops the work and pushes it into personal accounts. One-way redaction destroys the context the model needs, so the answer comes back unusable and people route around the control.

Salus takes the third path. Sensitive values are found inside your environment and replaced with typed, deterministic tokens: PHONE_qzvtrn instead of a phone number, PERSON_kxmrqa instead of a name. The tokenised request goes to the provider, the provider writes its answer around the tokens, and Salus puts the real values back inside your perimeter before the answer reaches the person who asked. The workflow is unchanged and the answer is complete.

Where the intelligence runs

The step that has to see real data, detection, runs on your infrastructure. Format rules with checksums handle structured identifiers such as TCKN, VKN, IBAN and card numbers. Language models handle names, organisations and locations. A second model verifies every model candidate in context before anything is replaced. None of this calls out.

The external model only ever does one job: reasoning and writing, over tokens.

What stays inside

The mapping from token to value is held in a PostgreSQL store that you run, under a key that you generate. Restoring a token is a separate, authenticated API call, scoped to the tenant that created it. There is no Salus-hosted service in the path, no telemetry, and no vendor access to the store.

Two ways to put it in the path

Integrated. Your application, your batch job or your AI gateway calls the Salus API: protect before the provider call, restore after it. If protection cannot complete, the request fails and nothing is sent. See Integrate.
Endpoint. For the AI tools your teams use directly, Salus is deployed on managed devices and protects what is typed or attached without changing the application. See Endpoints.

Both paths use the same engine, the same tokens and the same rules.

What this documentation covers

The integration surface (endpoints, request and response shapes, contexts, restore), the architecture of the engine (services, detection pipeline, the token vault), and how to deploy and operate it (OpenShift and Compose, configuration, the management console, audit and metrics). Every statement here is checked against the current release of the engine. Example values throughout are synthetic.