Skip to main content
    English chalk downland at dawn — low golden mist in the valleys with a warm amber sunrise.

    Sovereign AI for schools

    Give teachers their evenings back.On data that never leaves the building.

    Teacher Manager marks pupil work and drafts feedback your teachers approve.

    Training on the UKRI Dawn supercomputer (Cambridge)

    AIRR national compute

    Member, AIRR User Advisory Group (UKRI)

    AI-generated atmospheric imagery. Read our imagery policy

    The problem

    Why are teachers still working after the children go home? Marking and paperwork eat the week.

    UK teachers work around 50 hours a week, and the biggest drains outside the classroom are marking, planning and reports — the work that follows a teacher home. That is the week Teacher Manager is built to give back: the drafting and the first pass, done in the building, so the teacher's evening is spent teaching tomorrow, not typing tonight.

    Source: Schools Week — teacher workload survey

    What it is

    What is Aevum Teacher Manager? One system for the whole school.

    Teacher Manager is an integrated suite, not a single tool: marking and feedback lead, and the office work, planning, leadership admin and tutoring that depend on them sit in the same system. None of it is handed to someone else's cloud: your data stays in the building, with you.

    01
    Model trained

    Marking & feedback

    The lead. Teacher Manager marks pupil work against the mark scheme and drafts WWW/EBI feedback the teacher reviews, edits and approves. The teacher stays in the loop — it is never a sole marker.

    02
    In integration

    School paperwork

    Reports and parent letters (with EAL versions), SENCO paperwork — IEPs, EHCP contributions and inclusion strategies — attendance and EBSNA letters, and Pupil Premium statements, drafted from the records already in the building.

    03
    In integration

    Teaching & planning

    Differentiated lesson packs and resources for the classes you actually teach, with multimedia — images, diagrams and video — built in, licence-clean and provenance-recorded.

    04
    Designed

    Leadership & admin

    Role-aware help for the people who carry the school: SENCOs, heads of year, headteachers, business managers and MAT/trust roles — each seeing the work their role owns.

    05
    Designed

    Observation, CPD & appraisal

    Observation notes, CPD records and appraisal cycles kept in one place, drafted with the teacher rather than done to them.

    06
    Model trained

    Learn — the tutor module

    A school-controlled tutor for after-school and intervention sessions that targets the exact gaps marking reveals. SEND-first, and it does the hard subjects — maths and science, not just text.

    The models are trained and evaluated; what we're building now is the integration into the tools teachers actually use. Each module above shows where it sits on that path.

    Designed— specified and scoped, not yet built.Model trained— the model exists and has been evaluated; we are building it into the interface.In integration— being built into the tool a teacher actually uses.Integrated— in the product and working.

    One suite, one licence, nothing leaves the building.

    The difference

    Why Aevum and not another AI tool? Own the model. Keep the data. Inspect the reasoning.

    Most classroom AI is a website run from someone else's cloud. Teacher Manager is the opposite arrangement: the model belongs to the school, the data stays in the building, and the reasoning behind every mark is open to inspection.

    Data never leaves the building.

    Sovereign by design: pupil work, records and feedback stay in the building. Nothing you enter is sent to us, and nothing is used to train anything.

    SEND-first.

    Built from first principles for diverse access needs — and for everyone else. Accessibility is a gate, not a feature: WCAG 2.2 AA, built to conform.

    Does the hard subjects.

    Maths and science, with working shown and marked against the scheme — not just text-heavy subjects other tools avoid.

    Inspect the reasoning.

    Every mark and every answer is grounded in source, with the reasoning visible — so a teacher can check it, not just trust it.

    Early evidence

    How do we know it works? We measure before we claim.

    Every claim below is followed by what it means for your school, in plain terms. Where a study is still running, the number stays unpublished until it exists.

    Does the marking agree with the mark scheme?

    We are measuring that now, on real UK mark schemes, and we will not publish a number until people have validated it.

    What this means for your school: you won't be sold a figure. The teacher stays the marker of record and reviews every mark, and when we can show how closely it matches, you will see the number and how we got it.

    Every answer is grounded in a source.

    A mark or an answer points back to the material it came from, and the reasoning is visible to the teacher who signs it off.

    What this means for your school: a teacher can see why a mark was given and overrule it in seconds. No black box in a child's record.

    Built and tested for child safety.

    Every model release is tested against public child-safety benchmarks, including deliberately broken control models, because a test that cannot catch a model we broke on purpose is not testing anything.

    What this means for your school: safety is a gate a release has to pass before it ships, not a promise added at the end.

    Your school's model never leaves your building.

    Each school's model stays in the building, never pooled with other schools, never sent to us. We also test our own models with a published privacy attack, and anything that shows a signal does not ship as it is.

    What this means for your school: there is no shared pool for an outsider to attack, and the model itself is checked so it cannot memorise and leak a child's data.

    Trained on UK national infrastructure.

    Base models trained on Dawn, the AIRR supercomputer at the University of Cambridge, under UK law, on licence-clean data.

    Read: Trained on Dawn →

    What this means for your school: the model your pupils' work meets was built in the UK, on national research infrastructure, not on a US cloud account.

    The claims register

    Every claim, next to the way to check it.

    If a claim ever stops being true, it comes off this page. That is what the register is for.

    “Nothing you type leaves the machine”

    Your data stays in the building and is never sent to us. Never used to train anything. Memory visible and erasable, with a Remove that means a later search finds nothing.

    The data-flow note →

    “Accessibility is the quality bar”

    SEND appropriateness is a must-pass criterion, not a toggle. Stated provision and published criteria, documented on the Trust pages.

    The Trust Hub →

    “Trained on UK national infrastructure”

    Dawn, University of Cambridge, part of AIRR. Aevum trains its models on UK national research infrastructure.

    How it's built →

    “Built by a teacher”

    Twenty years of IT and Computer Science teaching, first in Portugal, then in England, primary through A-Level.

    The founder, in full →

    “We publish the gaps too”

    Thirteen DfE product safety standards, mapped in public, including the ones we do not yet meet.

    The standards mapping →

    Founding schools

    Shape it with us.

    A small number of founding schools join a co-design pilot: 50% off the first year, direct influence over what gets built next, and no long lock-in — a one-year term you can walk away from at renewal, never a multi-year contract signed on a promise.

    The founding cohort is forming now. We don't publish school names without permission — there are no logos here yet, and there won't be until a school asks us to add theirs.

    Become a founding school

    Trust, in writing

    What can a school hold us to? The commitments are published.

    Everything below is written down in the Trust Hub — the safety, data and accessibility commitments a school's data protection officer can read before a single conversation.

    Sovereign by design

    Your data stays in the building and never comes to us. There is no pupil-data sub-processor, because the data does not leave.

    UK GDPR, DPIA-ready

    A published data protection impact assessment, a data processing agreement in preparation, and the documents a DPO asks for, in one place.

    Safeguarding by design

    No child accounts on the open internet, no advertising, no dark patterns. Risks removed by design rather than managed by policy.

    Accessible as a gate

    WCAG 2.2 AA, built to conform — accessibility signs off a release, it does not trail behind it.

    Aevum Foundry AI is a private British company. Nothing on this page implies government endorsement.

    Contact

    Talk to the founder.

    One email reaches the person building it. Tell us about your school, and we'll tell you honestly whether Teacher Manager is ready for it.