Flows runs the routine work across your whole desk. AI drafts a flow from plain words, answers what you ask in Search or ask, and acts on its own only once your people have agreed with it.
Say what should happen, the way you would tell a colleague. Plugboard drafts the flow from your own tickets, repairs and settings, and asks first when it needs to.
A draft changes nothing. Publishing needs someone who holds every permission the flow uses.
“When a staff member joins, set up their account, licences and groups.”
WhenA staff member joinsTrigger
Only ifStart date and manager knownWaits
ThenOne go-ahead, with step-upA person
ThenAccount, sync, licences and groups6 steps
Drafted from your school’s account templates.
Illustrative draft. Nothing runs until it is published.
Try it before it touches anything.
Run the draft on a real past ticket or person and read what it would do, step by step. Backtest it over recent records, then publish it in Watching mode to record what it would have done.
Would ask for one go-ahead with step-up, covering the next 6 steps.
Would build Priya Sharma’s account from the Staff template.
Would add her to the staff groups in Active Directory.
Would start a directory sync and wait for her cloud account.
Thomas Walsh: would wait for a start date and manager.
Read it, change it, try again.
Illustrative dry run on sample people.
Decisions that earn their automation.
A Decide step asks one typed question, such as which kind of request this is, and follows the lane for the answer. Each answer is Suggested until a person confirms it, and the question acts on its own only once people have agreed with it often enough.
Unsure, unanswered or below your threshold, the run asks a person.
Is the requester asking for access for someone else?
AnswerYesSuggested
ConfidenceAbove your thresholdChecked
ModePeople confirm each answerSuggest
Confirm the answer
Only a person’s choice counts toward the question’s record.
Illustrative decision. Automatic outcomes never count toward its record.
Guard rails on every step.
Changes to Plugboard’s own records run on their own. Changes in other systems wait for a person’s go-ahead, and accounts, devices and admin roles need step-up too. Every run keeps a timeline, and a failed step pauses with Retry, Skip and Undo.
Nothing is deleted, and a run never acts with more access than a person.
AccountBuilt from the Staff template in Active DirectoryDone
SyncEntra Connect SyncDone
LicencesMicrosoft 365 licences and groupsRunning
Undo turns an account off. It never deletes.
Illustrative run on sample records.
Choose a stage to follow the flow.01 / 04
The detail, when you need it.
What flows can reach, the AI built into the desk, what stays under your control, and where your data goes.
What a flow can do
One builder for automation across the whole desk: tickets, repairs, loans, stock, visitors, kiosks, devices, knowledge and people, and the systems you connect.
When. A ticket or repair arrives or changes, a person joins, moves or leaves, someone asks for access, a schedule comes round, another app calls in, or a connected system raises an event such as a security alert.
Then. Change records, send messages, ask for approval, hand a task to a technician, wait, branch, decide, write with AI, repeat for each item, or call another flow.
People flows. Joiners, movers and leavers built from your account templates: Active Directory, a directory sync, licences, groups, mailbox delegation, and device and loan recovery, with one go-ahead and step-up, and an undo that turns accounts off and never deletes them.
Privileged access. Entra PIM roles activated by the person themselves, and just-in-time admin inside Plugboard that someone else approves and that ends on time.
Approvals anywhere. Put an approval in front of any action. A flow can also offer itself on a ticket, repair or person, as a card a technician runs, sends for approval or dismisses.
Other platforms. Hand a step to Power Automate, Zapier, Make or n8n through a signed connection, and start a flow from another app through its own hook.
How decisions earn their automation
A decision is a typed question with a closed set of answers: yes or no, one of a list, or a score. The answer chooses a lane in the flow. It never chooses who or what a step changes.
Suggested first. Each answer is shown to a person, who confirms it or picks another. Only those human answers count toward the question’s record.
Earned, per question and per model. A question acts on its own only after your people have agreed with enough of its answers, and only if your school allows it. A few corrections send it back to suggest.
A fallback that always works. A condition written in words also has a keyword or field rule. With AI off, or no answer, the rule decides or a person does.
The model that answers. Managed schools get Clef, on Cloudflare Workers AI. A self-hosted school can run Laya on its own server, and any school can add TypeSafe Jev on its own TypeSafe account, hosted in the United States. Decision text is always pseudonymised.
The AI built into the desk, as flows you can open
Sorting incoming tickets, sending them to the right desk and queue, resolving on the spot, suggesting a device fix and spotting access requests run as system flows. Open one to see what it does, narrow it to one desk, or turn it off.
Each keeps its own mode, threshold and track record, inside your daily limit on background AI calls.
Routing and access requests learn from the answers your team corrected, inside your school only.
Email and Teams intake is read for the repair type and your school’s decision criteria, and acknowledgements and automatic replies are kept apart from messages that need someone.
When a requester says the fix worked, a technician confirms the resolution. The AI never closes a ticket quietly.
Each week Plugboard points out work your team keeps doing by hand that could be a flow, from your records alone and with no AI calls.
Raising a ticket in one sentence
Type what happened. The ticket and repair composers fill in the person, how it came in, the desk, the category, the priority and the device, and mark every field they filled so you can change it or undo it. They never write over a field you set.
Resolved on the spot. When the words say the job is done, the ticket is raised already resolved, with the fix kept as a note.
Polished wording. Turn quick notes into a clear description, or keep your own words.
@-mentions. Mention a person, device, ticket or repair to link it. Mention a colleague in an internal note and they are told, if they can see the ticket.
Needs you, Spotted and Search or ask
Home shows what needs you and what Plugboard spotted: a cluster of reports, a device that keeps coming back, a loan pool about to run short. Anything changed automatically overnight is in the activity with its reason and an Undo, or says why it cannot be undone.
Search or ask is the one place to ask. The answer opens in the Ask panel, signed Plugboard, with the records it came from. Ask “what does this person have on loan?” on a student’s page and the answer is about that student, straight from their loans, devices, repairs and tickets.
One ticket per problem. A same-person duplicate within 48 hours can merge on its own, with a one-click unmerge.
One incident per outage. When several people at one site report the same outage, Plugboard proposes one incident with the evidence. Accept it and the reports are grouped, a known issue is posted, and each requester is told once.
Repair outcomes and approved routine actions
At lodgement, a likely-outcome card shows the probable result and how many days it should take, drawn from your own repair history.
Cover agreements. A repair covered by the school’s insurance with its vendor never asks a family to approve or pay.
Replace device. Take the replacement from stock or the loan already out, or record one on its way, and say what happens to the old one. The repair, both devices and the person’s history record it.
Access requests arrive as a card that already knows the groups, the licences, the free seats and who approves.
Device fixes for staff devices come from a catalogue. The AI suggests one, a person approves it, and your MDM runs it.
Tone, urgency and safeguarding
On staff tickets, the AI reads each message for tone and urgency, and for a possible welfare concern for staff to review. The queue has a Tone column and a Frustrated requesters filter. With AI off, a rules pass on chases, reopens and waiting time gives a coarse read, labelled as rules.
Tone never changes a person’s account, access or charges. In auto mode it can raise a staff ticket’s priority one step, never to the top, and that can be undone.
Staff get an experience score: a 90-day summary of waits, reopens and satisfaction on the tickets they raised. It is a conversation starter for a manager, not a performance measure.
No tone or welfare reading for students and parents
In every mode, shadow included, the AI does not read how a student or parent sounds, or look for welfare concerns in what they write. No score is kept about them, and their experience appears only in school totals. The safeguarding word list stays on: it is a list of words, not AI, and a match shows the school’s own contact.
What runs automatically, and what needs approval?
The AI proposes. People and your records decide.
Labelled in words. A Suggested line sits on the record it is about, a Draft waits in the reply box, and a Summary opens with the messages it was read from. Anything set automatically says so, with its reason and Undo, and says when something, such as a device restart, cannot be taken back.
Off, shadow, suggest, auto. Set per feature, and per question inside a flow. Shadow compares proposals with what staff actually did. Auto unlocks only after people have agreed with it often enough.
Never automatic. Accounts, devices and admin roles always wait for a person’s go-ahead with step-up, inside a flow or anywhere else.
Writes wait for a person. A write requested through an API key or MCP is held until someone with the right permissions approves it.
Every call on record. Admin › AI activity lists every model call and suggestion, searchable and exportable. Calls are logged as counts, not prompt text.
Nothing generated reaches students or parents. They get help articles found by search and translations a person has approved.
Off means gone. Switch AI off for the school and none of it is drawn: no Draft button, no empty box, nothing greyed out.
Where does AI run, and what leaves our school?
Your hosting region and where AI text is processed are separate answers. This is the AI one, for each way a school can run Plugboard.
Managed AI
Writing and answers use Claude, from Anthropic, through Cloudflare, processed by Anthropic in the United States whatever your hosting region. Similar tickets (BGE-M3) and decisions (Clef) run on Cloudflare Workers AI. Names, emails, phone numbers and IDs are pseudonymised before every call.
AI sub-processors: Cloudflare, and Anthropic in the United States.
Self-hosted, local model
The bundled local model runs on your own server. The text stays on your network.
No AI sub-processor to disclose.
AI off
Switch it off for the school and nothing is sent. Triage runs on rules, Decide steps ask a person, and nothing that needs AI is shown.
Nothing leaves for AI.
A school can also connect its own OpenAI, Anthropic or compatible account. That provider works under the school’s contract, and pseudonymisation stays on by default.
What pseudonymisation does not catch: a description that identifies a student without naming them, such as “the kid in 9B with the red case”, or a name that is not on your roster. The data-processing page lists these gaps and the sub-processors.
Cost and other questions
Is this a chatbot?
No. You can ask in Search or ask, but most of the work is prepared in the background and shown where you already work: on the ticket, in the queue, and in Needs you and Spotted on Home.
Does it act on its own?
Only where you set a feature or a question to auto, and auto unlocks only after people have agreed with it often enough. Accounts, devices and admin roles never change without a person’s go-ahead, and no reply goes to a requester without a person.
Do we need AI to use Flows?
No. Flows run on their rules and their people. With AI off, conditions written in words use their keyword rules, Decide steps ask a person, and you build each flow step by step.
Do students or parents talk to the AI?
No. No generated text reaches a student or parent, and the AI does not read their tone or welfare. They get help articles found by search, known issues and translations a person has approved.
What does it cost?
A monthly allowance of hosted AI calls comes with the licence, and extra calls are charged at the rate in your quote. A local model or your own key costs nothing from us. How pricing works
What if we don’t want AI?
Turn it off for the school, or per feature. The desk runs on its rules, nothing is sent anywhere, and nothing that needs AI is shown.
See it on your own tickets.
Fifteen minutes on your queue, your student system and the work you would hand to a flow first. A 30-day trial on your own school’s data follows the demo.