Outbound only. Read only. No inbound ports.
IntraLogic reads infrastructure to build a model of it. It is not a management tool, and it is not RMM — there is no remote shell, no remote desktop, and no write path into your environment. This page states the mechanism behind each of those claims, and the places where the boundary is not as clean as we would like.
Configuration and inventory metadata — hostnames, models, serials, capacity, interfaces, dependencies. No file contents, no databases, no user data.
Read-only. A full compromise of IntraLogic could disclose how your environment is built. It could not change it.
IntraLogic is in preview and publishes no SLA yet. Discovery is asynchronous — if the platform is unreachable, your infrastructure is unaffected.
The collector never accepts a connection.
Discovery runs from an agent inside your network. It polls the IntraLogic backend over outbound HTTPS, does the work, and posts results back. Nothing connects inward to it.
No inbound ports
The agent opens no listening port to the network. There is no firewall rule to add, no NAT entry, and no VPN requirement.
Loopback-only diagnostics
The local troubleshooting endpoint binds to 127.0.0.1. It is not reachable from another machine on the same LAN.
Not remote access
No remote shell, no remote desktop, no file transfer, no script delivery. The agent runs only the discovery code it shipped with.
What each integration asks for, and what it cannot do with it.
“Read-only” is only meaningful per integration, so here it is per integration. Every connector is restricted to retrieval calls; where a vendor makes that impossible, it says so in the last column rather than being left out of the table.
| Integration | Credential | Access it uses | What it cannot do |
|---|---|---|---|
| VMware vCenter | Account with a read-only role | vSphere Automation REST session, inventory GET requests, then logout | Power on/off, reconfigure, snapshot or migrate — no write call exists in the connector |
| Microsoft Hyper-V | Windows account permitted to use WinRM | Bundled Get-* / CIM scripts run on the target | Run any mutating cmdlet — blocked before execution, see enforcement |
| TrueNAS | API key, or user and password | REST v2.0 GET requests; an optional SSH pass for pool detail | Change configuration — the SSH commands are allow-listed too |
| Proxmox VE | API token, or a login ticket | /api2/json GET requests | Write config; no shell access is used |
| Nutanix Prism Central | Account with Basic auth | v3 list queries — a search, not a change | Any mutating call; no Controller-VM shell |
| Dell iDRAC & Redfish | BMC account; read access is sufficient | Session token, then GET requests, then the session is deleted | PATCH or any Redfish Action — never issued |
| Cisco IOS-XE | RESTCONF account over HTTPS | RESTCONF GET requests against YANG models | edit-config — no configuration write |
| Cisco NX-OS / IOS | SSH account | show commands only | Enter config mode, copy, write, reload — see enforcement |
| SNMP generic, UniFi, Aruba, Dell | v2c community string, or a v3 USM user | GET of standard OIDs | SET — the module implements no write operation at all |
| SonicWall | Full administrator | GET requests after an auth POST | The connector issues only reads — but SonicOS has no read-only API role, so the credential itself is privileged. Stated in full below. |
SNMP v3 uses USM with SHA-family authentication and AES privacy where the device supports it. Weaker options (MD5, DES) are accepted because production switches still require them — they are the device's limit, not a default we chose.
Not “we don’t” — the code refuses.
Read-only by convention lasts until someone edits a file. Both CLI-based connectors put a guard between the discovery code and the transport: every command is inspected before it is sent, and a violation aborts the run rather than executing.
PowerShell & CIM (Hyper-V)
Scripts are scanned for mutating verbs at a token boundary. A match throws before the transport opens.
Every cmdlet family that creates, changes, removes or power-cycles anything on the target is blocked. A small number of cmdlets that only open and close the local session are permitted, because they never touch the target. Matching is on the whole cmdlet name, so a permitted prefix does not carry a mutating cmdlet through with it.
SSH CLI (Cisco and others)
Two layers, and a command must pass both. First, the leading token must be on a short allow-list of read-only verbs — the handful a switch uses to report state and nothing else.
Second, a banned-token scan rejects any line containing a state-changing keyword anywhere in it: configuration entry, file writes, erasure, reloads, interface state, and shell escapes. No user-supplied command is ever run — the set is fixed in the agent. Separately, if a device unexpectedly drops into a configuration prompt, the run aborts immediately.
A permission granted to a connector is only as safe as the code holding it. These guards mean the boundary survives a future edit — a mutating command added by mistake fails the discovery run instead of reaching your device. The exact verb and token lists are shared in full during a security review — they are a defence against our own future mistakes, not a secret, but there is no reason to publish a bypass checklist.
How the secrets you enter are stored.
AES-256-GCM at rest
Secrets are encrypted with AES-256-GCM before they reach the database. The application refuses to start if the encryption key is missing or the wrong length, so it cannot silently fall back to storing plaintext.
Write-only through the API
A secret can be set or replaced. No endpoint returns one. Re-assigning a credential copies the ciphertext directly rather than decrypting and re-encrypting it.
Never logged
Community strings, keys and passwords are passed to connectors in the job payload and kept out of logs and error text. Hyper-V credentials reach PowerShell through environment variables, never the command line, where another process could read them.
Three places the boundary is imperfect.
Every one of these is a real trade-off we made, not an oversight. They belong on this page rather than in a support ticket after you have deployed.
SonicWall needs a full-admin credential
SonicOS provides no read-only API role. The connector issues only GET requests, and the read-only guarantee for SonicWall rests on that code — not on the permissions of the account. If that is not acceptable in your environment, leave the SonicWall integration unconfigured; nothing else depends on it.
TLS verification can be turned off per connection
Baseboard controllers and appliances ship with self-signed certificates, so each connection can opt out of certificate validation. It is a per-connection choice, not a global default, and it applies only to that one endpoint — but where it is enabled, that session is not protected against an on-path attacker inside your network.
SSH host keys have a permissive mode
Three modes: strict pins an expected fingerprint and fails on any mismatch; trust-on-first-use records the key on first contact and refuses if it later changes; compat accepts a changed key. Validation is never disabled globally, and a changed key is always recorded and reported for review even in compat mode — but in that mode the connection proceeds before you have seen it.
Nothing enters the model without review.
Discovery proposes; people accept. What was found, what was changed, who approved it and which version it produced are all recorded.
Review before adoption
Detected differences are presented for confirmation. A discovery run does not silently rewrite the model your team is working from.
Published baselines
A reviewed model becomes the approved baseline for that client, and new discoveries are compared against it rather than overwriting it.
Audit trail
Actions record who changed what, when, and from which source, so a model version can be tied back to the evidence and the event it came from.
Role-aware access
Access is scoped by organisation, client, module and role, so an engineer sees the environments they are responsible for.
Multi-factor authentication
Required by default, with two methods: a six-digit code sent by email, or an authenticator app using standard time-based one-time passwords. There is no SMS option — SIM-swap is a real attack and we would rather not offer the weakest factor at all. Only a super administrator can exempt an account; a tenant administrator cannot switch it off, for themselves or anyone else. Authenticator secrets are encrypted at rest, and reset links and emailed codes are single-use and time-limited.
Every sign-in is on the record
Each attempt is stored with its source IP and user agent, whether MFA was presented and whether it passed, which flow it came through — password, MFA, or password reset — and, when it failed, why. Users can review their own history from their profile without asking anyone. Accounts can be enabled and disabled, and the record keeps who did it and when.
Tenant isolation, and where it is enforced
Each client is a separate environment — model, discovery, credentials, documents, reports and event history. That boundary is enforced in the application layer: every request is scoped to the caller’s client before it reaches the data. It is not database row-level security, and we would rather name the layer than let “isolated” imply one further down than it is.
Live and Simulation separated
Simulation runs against the model without touching the live operational view, so a what-if can never be mistaken for a real state change.
AI is review-first
The assistant reads the model and proposes; a permission-checked, user-confirmed deletion path is the one write it can request, and it requires your approval.
What we can and cannot claim today.
No SOC 2 report yet
IntraLogic holds no SOC 2 report today, and saying otherwise during preview would be worth nothing to you. When one exists we will name its type and its observation period. There is no such thing as a SOC 2 certificate — it is an attestation issued by an auditor, and any vendor claiming to be “SOC 2 certified” is describing something that does not exist.
GDPR and CCPA are laws, not badges
They are obligations that apply to how we handle personal data, not certifications to display. IntraLogic collects infrastructure metadata rather than personal data; the personal data we do hold is account information, described in the privacy policy.
Ask for whatever this page does not answer — questionnaire responses, architecture detail, or the specific permissions for one integration. Get in touch and we will answer directly rather than pointing at a badge.