Every mechanism, written down.
Where the data lives, who can reach it, and how every report from every agent is proven before it is accepted. Written for the technical lead who has to answer the client's security questionnaire.
Last reviewed 20 October 2026
Hosted in the Netherlands
Client data is stored and processed on Secuno's own servers in the Netherlands, and does not leave the EU.
No outside AI, ever
The AI engine runs on the same servers. Inventories, CVE data and risk records are never sent to OpenAI, Anthropic, Google or any other AI provider.
Every client sealed off
Tenant separation is enforced by the database itself, underneath the application's own checks.
Every action on the record
72 audited action types, with who, when, from where, and the old and new value. Logs cannot be deleted from the interface.
What the platform will not do
Some protections are not settings. They are paths that do not exist.
- NEVER
Load an unsigned rule set
Detection rules are Ed25519-signed and verified before they load. A build without the public key accepts no rule set at all. An unsigned path would be a remote-execution channel into every agent, so there is none.
- NEVER
Run a command outside the allow-list
The server checks every response command against eighteen allowed types before it is queued, and signs the order once, with an expiry, so it runs only once. A process kill carries the image and start time the detection saw, so it cannot hit a recycled process ID. Kills never hit critical system processes, and protected or boot-critical services cannot be stopped or disabled.
- NEVER
Strand an isolated device
Network isolation is installed in one transaction with the agent's own exception, and expires locally unless a heartbeat renews it. An unreachable device rejoins the network instead of being cut off with no way back.
- NEVER
Let support act unseen
Support can step into a user's session for at most thirty minutes, under a banner that cannot be dismissed. Every action is logged as the admin acting on the user's behalf.
- NEVER
Trust an IP address a client supplies
A device's public address is used for display, location and audit context only, never for an authorisation decision, because the client supplies the headers it comes from. Sign-in rules such as an allowed-countries list read the address the platform's own edge establishes from trusted hops.
- NEVER
Put a token in a URL
No session or sign-in token ever appears in a URL, there is no self-registration, and enrolment tokens are shown once and stored only as a hash.
- NEVER
Block a device's own lifelines
An indicator block can never target the platform's own addresses, known remote-management vendors or a customer's gateway. On the device a block never covers the gateway or DNS servers, and it lapses when the platform stops renewing it.
- NEVER
Ask a user for a password
Restart prompts carry no links and always say that Secuno never asks for a password. Secrets in command lines and scripts are redacted on the device before anything is uploaded.
One report, proven five ways
Every request an endpoint agent sends is signed by that device and checked before a single field is read. This is what happens to one inventory report.
At enrolment the agent creates its own Ed25519 key pair. The private key never leaves the device and is protected with Windows DPAPI under the agent's service account, so another local account cannot decrypt it. The key is rotated after 30 days, with the new key signed by the old one.
The agent hashes the complete request body with SHA3-256. Change one byte and the hash no longer matches.
Method, path, client, device, timestamp, a random nonce and the body hash are signed together with the device's private key. A field cannot be moved or swapped without breaking the signature.
The agent connects over TLS 1.3 unless an administrator explicitly allows 1.2. Because the report itself is signed, a network in between cannot alter it unnoticed, even where TLS is terminated before our servers.
The platform verifies the signature against that device's public key, recomputes the body hash, rejects a timestamp more than five minutes off, and refuses a nonce it has seen before. Only then is the report processed, scoped to the device's own client.
Three locks, not one
An MSP platform holds many clients side by side. A single forgotten filter must never be enough to show one client's data to another, so separation is enforced three times.
- 01 · Application
Checked on every route
Each request is checked for the tenant it belongs to before any data is touched.
- 02 · Database
Row-level security underneath
The application connects with an unprivileged role. A query that forgets its tenant filter returns nothing, and a row cannot be written for a tenant the request has no business with.
- 03 · Build
No new table without a policy
A build check fails when a table holding client data is added without an isolation policy, so the guarantee does not wear down as the product grows.
Enforced in production
Row-level security is enforced in production on every client-owned table. A query that forgets its tenant filter returns nothing, raw SQL included. Only the cross-tenant operations role is exempt, and the CVE and AI services can reach the CVE catalogue and its correlation data, and nothing else. Enforced since September 2026.
Not found, instead of forbidden
A request for another tenant's record gets 'not found', so the platform does not even confirm that the record exists.
Agents stay inside their own client
An agent request is scoped to its own client and the partner that owns it, and nothing wider.
Who gets in, and how
Access to the console follows the same rules for an MSP technician and for the client's own staff.
- Sign-in
- Microsoft Entra ID or Google single sign-on, set per client: which providers, which tenant or Workspace domain, and whether a password remains. Microsoft sign-in requires a pinned Entra tenant, and a client can pin only the tenant its Microsoft 365 connection proves. Requiring SSO is a decision a client's own super user cannot take for their organisation.
- Second factor
- Authenticator app (TOTP), FIDO2 security keys and passkeys such as Windows Hello and Touch ID, up to ten named keys per user. Enforceable per role, per client and per user. A second factor enrolled in Secuno is proven after single sign-on too.
- Sessions
- Access tokens expire after 15 minutes, refresh tokens after 8 hours, and a session ends after a day however often it refreshes. Sign out everywhere ends every session a user holds on their next request, without blocking the account.
- Brute force
- Rate limits shared across all backend instances, and a lockout after 5 failed attempts for 15 minutes by default. A locked unknown address looks the same as a real one, so the lockout cannot reveal which accounts exist. A password reset takes the same time whether or not the address has an account.
- Roles
- Eleven roles, from Global Admin to Client User and a wall-screen display account, checked by the application and by the database underneath it. A display account sees only the dashboards shared with its screen, and never gets record lists, exports or writes. Two analyst roles for a managed-service team, L1 triage and L2 investigator, are assigned per client with two-factor mandatory by default, and a build check fails unless every permission is either granted to an analyst or written on a deny list.
- New users
- No self-registration. Users are invited, imported from CSV, or provisioned over SCIM 2.0 from Entra ID or Okta, and a SCIM deactivation ends their sessions at once. Invitations expire after 7 days and can be revoked.
- API keys
- A key acts within its owner's roles and client reach, narrowed further by scopes per area (read, read and change, or none) and optionally by client and IP range. Actions on endpoints, mailboxes, firewalls or access are never on by default. Never valid for sign-in, key management or SCIM, stored hashed, expiring within a year, rate-limited, and every change it makes records which key made it. A recognisable prefix and checksum let secret scanners catch a leaked key.
- Separation of duties
- Nobody approves their own decision. Accepting a risk is a proposal someone else approves, the author of an automation rule cannot approve that rule's runs, a remote script or uploaded software package needs a second person's approval before it can run, an emergency patch push needs a different approver, and wiping a device needs a second person within 24 hours. A client can also require that every device management change waits for someone other than its author. Wipes, locks and reveals of escrowed keys are refused when their audit record cannot be written. There is no admin exception.

AI that never leaves the building
SECAI, the platform's AI engine, is a locally hosted language model on Secuno's own servers in the Netherlands. It helps match CVEs to installed software, enriches risks and suggests fixes, but it does not decide on its own: the measured, deterministic score comes first, and the AI layer can only adjust it within a fixed bound.
- 01No inventory, CVE or risk data is sent to an outside AI provider
- 02Every input behind a score is visible in the console
- 03A control counts only against the attack path it actually blocks
- 04The AI adjustment is capped, so it can shade a score and never overturn it
- 05Behavioural baselines and attack stories run in the platform itself, not in the AI model, and client report narratives are written without an AI call
You do not have to take our word for it
These are the security headers the platform API sends on every response. Run the command and compare.
$ curl -sI https://app.secuno.nl/api/health
HTTP/2 200
strict-transport-security: max-age=15552000; includeSubDomains; preload
content-security-policy: default-src 'none'; script-src 'self'; … frame-ancestors 'none'
cross-origin-opener-policy: same-origin
cross-origin-resource-policy: same-origin
cross-origin-embedder-policy: require-corp
permissions-policy: camera=(), microphone=(), geolocation=(), payment=(), usb=(), …
referrer-policy: same-origin
x-frame-options: SAMEORIGIN
x-content-type-options: nosniff
Our security contact is published in security.txt (RFC 9116).
The answers clients ask for
NIS2 makes your clients check their suppliers, and that includes the platform you run their security on. These are the short answers; ask us for the detail behind any of them.
Where is data stored and processed?
On Secuno's own servers in the Netherlands. Client data does not leave the EU.
Is data shared with AI providers?
No. The AI engine runs on the same servers in the Netherlands.
How is data protected in transit?
HTTPS with HSTS preload for the console; agents use TLS 1.3 by default and sign every request with a per-device key.
How are tenants separated?
By application checks and by database row-level security, enforced in production since September 2026.
Which sign-in options are there?
Entra ID or Google SSO, TOTP, FIDO2 security keys and passkeys, enforceable per role, client and user. Microsoft sign-in is pinned to the client's own Entra tenant.
Is activity logged?
Yes, 72 action types with user, time, source and the old and new value. Logs cannot be deleted from the interface.
Can staff see what the agent does on their device?
Yes. Each organisation can publish a public page, in English or Dutch, that tells the people whose devices run the agent what it is, who operates it, what it can and cannot see, and who to ask. It shows no device names, findings or personal data. An optional Windows tray icon shows the same disclosure, with no actions and no channel into the agent.
Which subprocessors are involved?
Ask us for the current list, with what each one does and where. Request the list
How long is data kept, and how is it deleted?
Set out per data type in the data retention reference. Data retention
How do I report a vulnerability?
Email [email protected], in English or Dutch. The contact is also published in security.txt. [email protected]
Where can I see platform status?
On the live status page. System status
Every control, for the reviewer
The complete list behind the sections above, for the person who reads security pages line by line.
Endpoint agent11
- Per-device Ed25519 identity
- Created at enrolment, protected with Windows DPAPI under the agent's service account so another local account cannot decrypt it, and rotated after 30 days with the new key signed by the old one.
- Signed requests
- Every request is Ed25519-signed over method, path, client, device, timestamp, nonce and a SHA3-256 hash of the body. A signing input that tries to smuggle in an extra line is refused.
- Replay protection
- Timestamps must fall within five minutes, and every nonce is single-use.
- TLS 1.3 by default
- The agent requires TLS 1.3 unless an administrator explicitly allows 1.2, and does not skip certificate checks.
- Offline spool
- When the platform cannot be reached, reports are written to an on-disk spool of up to 64 MiB and replayed when the connection returns, so a network outage does not leave gaps.
- Single-use enrolment tokens
- Shown once in the console, stored only as a hash, with a configurable expiry and maximum number of uses.
- Signed rule sets, fail-closed
- Local detection rules are verified before they load; without a valid signature nothing loads.
- Allow-listed commands
- Eighteen response types: kill a process or its tree, isolate and release the network, apply and lift a lateral quarantine, quarantine a file, stop or disable a service, reboot or cancel a reboot, sync now, run a YARA sweep, uninstall one named version of an application, apply or revert a built-in vendor workaround, and the indicator-block channel. A workaround order carries only a catalogue id, never a registry path or a value.
- Signed orders
- The platform signs every order once, at queue time, with an expiry. Agents refuse unsigned or unverifiable orders, bind each one to their own device and client, and run it only once.
- Guard rails on the device
- Kills go by process identity and never hit critical system processes, boot-critical services cannot be disabled, each user account has its own budget of automatic responses, and secrets in command lines and scripts are redacted before upload.
- Signed replies
- On Windows agent 0.57 and later and Linux agent 0.49 and later, every platform reply the agent acts on is signed and bound to the request, the device and a nonce.
Console and accounts10
- Session lifetime
- 15-minute access tokens and 8-hour refresh tokens. The signing secret is checked at start-up, and the platform refuses to start with a weak one.
- Single sign-on
- Entra ID and Google per client. Microsoft sign-in requires a pinned Entra tenant, and a token from any other tenant is refused. A second factor enrolled in Secuno is proven after the provider sign-in too. Every refused sign-in is audited with its reason.
- API keys
- Act as their creator with current roles and client reach, read-only unless made otherwise, and never reach sign-in, key management, SCIM or credential resets. At most ten per person, expiring within a year, shown once and stored hashed. Minting one needs a recent sign-in and is refused while impersonating. Scopes per area and optional client and IP limits narrow a key further; keys unused for 90 days are revoked, deactivating a person revokes their keys, and use from a new country raises an alert. Owners are told 14 days, 3 days and on the day before a key expires, and each key may make 600 requests a minute. Integrations can use a service account instead, owned by the organisation rather than a person, with hour-long tokens that never carry more than the account holds at that moment.
- Two-factor
- TOTP, FIDO2 security keys and passkeys. Every challenge is single-use, so a captured response cannot be replayed.
- Rate limiting
- Shared counters across backend instances. Two-factor attempts are capped per temporary token, after which the user must sign in again. A fall back to local limiting is logged.
- Lockout
- 5 failed attempts, 15 minutes by default, both configurable, and indistinguishable for unknown addresses.
- Support access
- At most thirty minutes, under a permanent banner, audited as the admin acting for the user.
- User provisioning
- SCIM 2.0 from Entra ID or Okta creates, updates and deactivates client users, and a deactivation ends their sessions at once. Provisioning tokens are stored hashed, last one year, and stop working when their creator leaves.
- Live notifications
- The notification stream opens with a one-time ticket valid for 60 seconds. Only its hash is stored, so a ticket read from a log grants nothing, and no access token appears in the stream's URL. A stream closes within half a minute of its session being signed out.
- Display screens
- Wall-screen accounts pass a route allowlist on every request, load figures only for dashboards they can open, and never get record lists, exports or writes.
Web application5
- Content Security Policy
- A strict policy on every response that blocks inline scripts, unknown sources and framing.
- HSTS with preload
- Browsers are told to use HTTPS only, for all subdomains.
- Cross-origin isolation
- COOP, CORP and COEP on API responses, so a hostile page cannot hold a reference to a Secuno window or pull its resources into another context.
- Framing, referrer and permissions
- X-Frame-Options, a same-origin referrer policy, and a Permissions-Policy that switches off camera, microphone, location, payment and USB access.
- Parameterised queries
- All database access goes through parameterised SQL. No string building reaches the database.
Data and infrastructure5
- Hosting
- Secuno's own servers in the Netherlands. Client data does not leave the EU.
- Private AI
- SECAI runs on the same servers. No data is sent to an outside AI provider.
- Tenant isolation
- Application checks plus Postgres row-level security, enforced in production on every client-owned table since September 2026, guarded by a build check.
- Audit trail
- 72 action types with user, IP, user agent, location, entity, and the old and new value. Not deletable from the interface.
- IP addresses as context only
- A device's reported address is resolved in one place, used for display and audit, never for authorisation. Private addresses are refused rather than stored. Sign-in rules read the address the platform's edge establishes from trusted hops, never one the client supplies.
Found something? Tell us first.
Report a vulnerability in secuno.nl or the Secuno platform privately to [email protected], in English or Dutch. Include what you found and how to reproduce it. We confirm receipt, keep you informed while we fix it, and ask that you do not publish details until a fix is out.
Need answers for a client's questionnaire?
Send it to us. We answer security reviews and due-diligence questions, and walk you through any layer in detail.