No “enterprise plan” asterisk, no PDF-on-request. The boundary sits in the data layer, personal data is masked before any model sees it, and every assistant move happens in front of you. This page describes shipped behavior — each claim is testable during the trial.
Test every sentence on this page during the seven-day trial.
Most security pages are collections of badges. Ours is a list of behaviors. Where the badge would go, we describe what the code does.
The AI boundary is enforced before the prompt is built — not promised in a policy document afterwards.
Card numbers, IBANs, national IDs — 13 pattern families are detected and masked before any AI call. The checksums are validated, not guessed, and unmasking stays a permission, not a habit.
You choose where inference runs. EU-only is a setting, not a favor. Prompts do not wander across borders because a provider was briefly cheaper.
Your conversations train nothing — not our models, not a provider's by default. Bring your own provider keys and the boundary holds either way.
Arrives scoped
The message lands in your tenant. The data layer scopes every read and write to your workspace — isolation is enforced per query, not per promise.
Masked
Identifiers are detected and masked — with checksum validation — before the prompt is assembled. The model works on a redacted copy.
Pinned inference
The request goes to a model in your allowed region. You picked the region; the routing respects it.
Visible tools
If the assistant opens the order, checks the knowledge base, or browses the web, you watch it happen — tool calls render in the thread, not in a log you request later.
Who sends what
Who did what is not a feature you unlock — it is how the platform is wired. These ship with the floor, on every plan.
Sign in with your identity provider (OIDC or SAML) and enforce it for the whole workspace. BootDesk is also a public OIDC provider — your agents get tokens, discovery, and JWKS like any first-class client.
OIDC · SAML
provider · consumer
2FA for every human account, not an add-on for admins.
Granular permission groups, teams, and organizations — access follows function, and unmasking personal data is its own permission.
The parts of security that never trend but decide whether your data is yours.
Every attachment passes malware scanning before a human or a model touches it. Suspicious files quarantine; the ticket keeps its history.
GRUPO OITO OÜ, Tallinn, Estonia. The DPA is public from day one — not a negotiation. Estonian law, EU data handling as the default posture.
Cancel from the workspace. Export through the open API. Retention windows are yours to set — we would rather post the exit than manage a lock-in.
Something missing? Write to suporte@bootdesk.eu — an engineer answers, not a form.
Logged, with its cost
Token use, model, region: recorded per call. You can audit the AI's month like a phone bill — line by line.
OpenAPI documentation on every endpoint, plus an LLM-readable catalog. Your security team reads the same specs your integrations run on.
The browser agent and code execution are per-workspace opt-ins with a kill switch, and code runs against a host allowlist. Capability is a decision, not a default.
