Skip to content
plugboard.appRequest a demo

Security and privacy

What we process

The answers a privacy assessment asks for: what Plugboard holds, what leaves it, and who processes it, for a managed deployment and for one you host yourself.

A managed deployment in Sydney with AI on. Connectors and email you enable send their own data, listed under What connectors send.
CategoryExamplesWhere it comes from
PeopleNames, year levels, tutor groups, usernames, email addresses, student numbers, card numbersYour SIS and directory
DevicesSerials, models, assignment, lifecycle fieldsYour MDM, and records created here
WorkSubmissions, tickets, notes, statuses, outcomes, costsCreated in Plugboard
LoansWhich device, to whom, whenCreated in Plugboard
MoneyPurchases, damage charges, software contracts, free-text owners, seat entitlements and reminder recipientsCreated in Plugboard
Execution recordsAutomation runs and approvals, owner-scoped bulk device jobs and target outcomesGenerated from requested work
Staff accountsNames, emails, roles, MFA enrolmentCreated here, or SCIM
AuditWho did what, whenGenerated
ConfigurationModules, workflow, templates, encrypted connector credentialsCreated in Plugboard
Network client sightingsWhich access point or switch port a device was last seen on, and whenYour network equipment, only if you switch client tracking on

Plugboard does not provide dedicated fields for health data or ethnicity. Free-text fields and attachments can still contain sensitive information entered by a user; the platform cannot promise that no such information is held.

Free-text notes on submissions and tickets can contain anything a staff member types. A note explaining why a family cannot pay for a repair might touch on sensitive circumstances. Restrict who reads those notes with the ticket.internal permission, and tell your staff what belongs in a note.

What leaves depends on the connectors you enable and on your hosting, email and AI. A managed deployment gets the managed AI connector at first boot, so check the connector list rather than assume nothing leaves until somebody configures one.

CategoryWhat leaves
MDMDevice serial numbers, models, assignment and compliance state
SISStudent and staff names, year levels, identifiers and email addresses
DirectoryNames, usernames, email addresses and group membership
EmailMessage content and recipient addresses
SMSPhone numbers and message content
TicketingTicket subject, body and requester details
WarrantySerial numbers
Repair vendorSerial numbers, fault descriptions and contact details
SecurityDevice identifiers and application approval requests
CRMAccount and contact details
PrintingUsernames, card numbers and print balances
AITicket text and what staff ask in Search or ask. Where it goes depends on your AI connector: see where AI fits

By category, not vendor: swapping Jamf for Intune does not change the answer, and your compliance page lists only the connectors you enabled.

Self-hosted. Your school runs the application and database, and the connector, email, backup and AI providers you choose may process data. Licensing and the optional count-only usage report contact us. The bundled local model keeps AI on your own server.

Managed. Every sub-processor, and where each one works:

Sub-processorPurposeWhereCustomer data
DigitalOceanRuns your deployment’s server and database, and keeps weekly server backupsYour agreed region, such as SydneyYes
CloudflareIngress, access control and request observabilityCloudflare’s edge networkTraffic and request metadata; retention depends on the configured service
Cloudflare R2
A bucket per deployment
Encrypted offsite backup copiesOceania, best effort, for Australian deployments; the nearest location hint elsewhereYes, sealed with your deployment’s own key
Cloudflare Workers AI
and AI Gateway
Embeddings (BGE-M3) and typed decisions (Cloudflare Clef); relays chat to AnthropicCloudflare’s network, no region pinningPseudonymised ticket text and prompts
Anthropic
through Cloudflare
Chat and drafts, on Claude Haiku 5.5United StatesPseudonymised ticket text and prompts
Your email providerNotifications and remindersWhere that provider processes mailRecipient addresses and message content

Your region does not make all processing regional: connectors, email, AI and edge services have their own locations, so confirm them in your deployment agreement.

A deployment reports counts only to us: Technical and Business seats, managed devices, the number of active students synced from your SIS, enabled modules, version and deployment id. The student count is a number only, reported beside a per-student licence for the re-count at renewal, and is never billed on. The report carries no names, records, serials, credentials or content, and a self-hosted deployment can switch it off with USAGE_REPORTER_DISABLED=1 and export local snapshots instead; see licence & billing. Billing, support and contact records are our own business records, and can identify the people who administer or buy the service. The website privacy policy covers analytics and demo requests.

Support access. Plugboard support cannot reach a managed deployment until an administrator grants support access, for a set time and with a reason, under Settings, Compliance. Every support request is recorded in your audit log before it reads anything, and a grant can be ended at any time.

A managed deployment keeps three backup copies. Two stay in your region. The offsite copy is placed by a location hint, which Cloudflare treats as best effort.

CopyHeldKept
Nightly, 03:00On your deployment’s own server, in your region90 days
Weekly, whole serverDigitalOcean, in the server’s own datacentre4 weeks
Offsite
Sealed with your key first
Cloudflare R2, in a bucket of its own with the location hint oc: Oceania, best effortUntil the bucket is deleted, 30 days after decommissioning

Cloudflare offers no Australian jurisdiction for R2, so we say Oceania, best effort, and do not say the copy stays in Australia. Other regions take the nearest hint: apac for Singapore and Bangalore, weur for Frankfurt, Amsterdam and London, enam for Toronto and New York, wnam for San Francisco. When a deployment is decommissioned we delete its bucket after 30 days, or your agreed window, and email you to confirm.

Ticket text reaches a language model only through an AI connector, and the three kinds differ:

  • The bundled model runs inside your own deployment and sends nothing anywhere.
  • Anthropic and an OpenAI-compatible endpoint send it to whoever you point them at, on your own key.
  • Managed AI is set up at first boot on every managed deployment. Cloudflare runs embeddings and decisions on Workers AI; chat and drafts use Claude Haiku 5.5, which Cloudflare relays to Anthropic for processing in the United States.

With AI switched off, or no AI connector, Search or ask still gives exact answers from the desk’s own records, and nothing leaves.

Before a managed call leaves, names on your school’s roster, email addresses, phone numbers, IDs, device serials and street addresses are replaced with tokens, and the real values go back into the answer when it returns. See pseudonymisation.

The real values go back into the answer when it returns. Year and tutor group are not masked, and neither is a name missing from the roster or a description that points at a student without naming them.

This covers Plugboard’s own AI, not a third-party MCP client you connect yourself, which hands results to its own model, so scope its API keys. Settings, Compliance names the AI connector you configured.

Identity panes and the licence-position report fetch provider data on demand and do not persist sign-in history, MFA methods or licence inventories. The software register stores manually entered commercial terms and exact SKU mappings. Privileged action audit records retain the target, technician, reason and outcome. Exchange delegation uses an encrypted PFX credential and private temporary files removed after the subprocess exits. These are distinct from persisting a provider inventory.

Two things are inferred, both for staff to read, and both about staff only. See sentiment & staff experience for who can see each.

The tone of a staff member’s ticket. Read from the requester’s own messages on that ticket, never staff replies, and stored on the ticket as a tone, an urgency, a short list of signals and one line of reason. With AI configured, it comes from a separate typed decision, pseudonymised the same way as every other call. Without AI, rules read it from facts the desk already holds, and it is labelled Rules. It is never shown to the requester, in the portal, to parents or at a kiosk. It lives on the ticket, so it is deleted with the ticket under your retention policy.

A staff member’s experience of the desk. A nightly 90-day summary, kept for staff only: median time to first reply, reopen rate, chases per ticket, average satisfaction, the share of tickets that read as frustrated, and a score computed from those numbers. With AI on, one sentence may be added for a staff member having a rough time, written from the numbers alone. No names and no ticket text are sent. A row not refreshed for 7 days is deleted, which covers somebody who has left or changed role.

Nothing about a student’s or parent’s feelings is inferred. A ticket from a student, a parent or a sender who matches nobody gets no tone, from the AI or the rules, and no AI welfare read, in any mode, shadow included, and nothing about it is sent to a model for either. The Experience report counts frustration over staff tickets only.

Safeguarding. Staff who can read internal notes are flagged when a student’s ticket uses a word on the school’s safeguarding list, which is a word match and not AI, and when the AI reads a possible welfare concern on a staff member’s ticket. A flag decides nothing; a person follows the school’s process.

Erasing a person clears the sentiment on tickets they raised, deletes the tone and welfare readings of those tickets, and deletes their staff experience row. A staff member’s subject-access export includes their staff experience row.

Off unless you turn them on, under Settings, Security. They are the one category in the table above that is location information, and most of the devices on a school network belong to students, so the decision is deliberately yours to make rather than something that arrives switched on with a connector.

With it on, Plugboard keeps one row per device: the access point or switch port it last connected to, and when. It does not keep a history of where it has been. Rows older than your window are deleted automatically; the window is between 1 and 90 days, defaults to 7, and there is no “keep forever” setting for this.

Sightings appear in a subject-access export and are removed by an erasure, the same as everything else in the table above. They are never shown on the student or parent portal.

Retention is configurable, and erasure is built in rather than a manual database edit: see audit and retention. Settings, Backups exports a portable copy of your own school at any time, without asking us: see backups in the app.