• Local-first AI for classrooms, workshops and meeting rooms
  • Made in Germany
  • AI for your lessons — on-premises, not in the cloud

The classroom AI that stays in the classroom.

You give it the books, the slides and the worksheets. Then it takes part in the lesson and learns along like a very diligent pupil — what was taught, how it was explained, and where the class actually got to. Transcripts, documents and memory stay on a machine in the school, not in a vendor cloud.

  • A brain of its own for every class
  • On-premises Central Brain
  • Anonymous speakers
  • Built for school operators under GDPR and FERPA
Cutaway of a school wing: a small listening box on a classroom window-sill, speech turning into labelled cards that stack in a cupboard marked Central Brain, with the cloud left outside the wall.cloudClassroomCentral Braintopicexplanationhomework

Why lessons leak

  • Teachers repeat themselves

    The same explanation, five times a week, because nothing in the room remembers what was already said.

  • Absent students lose the thread

    A week off, and the only way back in is somebody else's handwriting.

  • “AI in school” usually means a US API

    Most classroom assistants ship the voices of minors to a service you cannot inspect.

  • You can have the memory without the leak.

How it works

  1. A Room Brain takes part in the lesson

    Not a recorder — a passive participant. A small box with a microphone that learns along like a very diligent pupil: what was taught, how it was explained, and what the class has already covered. An LED shows listening, muted or answering, and the mute is physical, not a setting in an app.

    A small Raspberry-Pi-class box with a microphone and a pulsing LED.
  2. A Central Brain thinks on the school LAN

    One strong local machine — a Mac mini or similar, in a locked cupboard — transcribes, embeds and compiles memory. Every class gets a brain of its own, shaped by that class and that teacher. No hosted LLM, no hosted speech-to-text.

    A small computer in a locked cupboard with a LAN cable and no route to the internet.LAN
  3. Teachers add the material

    PDFs, slides, notes. Students and teachers ask at the box, or later in the school portal on the LAN or through the school's VPN or tunnel.

    A teacher dropping a PDF into a tray where it becomes labelled cards.chunk 1chunk 2chunk 3
  4. Memory evolves instead of piling up

    A changed homework supersedes the old one, so “what is due now?” and “what was assigned on Monday?” both answer correctly.

    A card superseding an older card, the older one greyed but still readable.homeworksupersedeshomeworkMondaynow

What we promise about your lessons

These are product guarantees, stated the way they are actually implemented — including the caveats.

  • No cloud AI for lessons

    Speech-to-text, embeddings, retrieval and answers run on the school's Central Brain. We do not send transcripts, documents, memory or sources to hosted LLM APIs.

  • Audio is a stream

    The Room Brain never writes audio to disk. The Central Brain does not keep audio in normal operation. During a supervised pilot an optional on-premises audio archive can be switched on for quality analysis, and is then purged.

  • Room isolation

    Every query is scoped to a room. By default each room only sees its own memory. An operator can explicitly link rooms — for example workshop groups reading a shared plenum — or assign shared document collections. There is no automatic cross-room visibility.

  • No student identification

    Speaker labels are anonymous: Teacher, Speaker-2, Speaker-3.

  • You can pull the WAN cable

    Day-to-day teaching does not need the internet. Optional internet is for updates, model downloads, remote administration, and — only if a school explicitly enables it — cloud text-to-speech of the final spoken answer text.

  • Retention and deletion

    Transcripts default to 30 days; memory is longer and configurable. Teachers and admins can purge.

What it actually gets used for

  • “What did I miss this week?”

    A pupil was off sick. They ask their own class’s brain and get back what was actually taught — not somebody else’s handwriting.

  • “How was that explained again?”

    Three weeks later, a student gets the explanation their teacher gave, in the words the class already knows.

  • A substitute teacher, up to speed in a minute

    Where the class got to, what has been covered, what is still open — without the handover meeting that never happened.

  • Questions during the lesson, from a tablet

    Students look up an earlier lesson themselves instead of the teacher explaining it for the fifth time.

  • Workshops, meeting rooms, support teaching

    Same box, different furniture. Independent group rooms, plus an explicitly linked shared plenum when the operator wants one.

  • Multilingual classrooms

    Lessons in German, English and Hebrew — a pupil can ask in the language they think in.

Who it is for

  • Teachers
  • School leaders
  • School IT
  • Districts and Schulträger
  • Workshop and academy operators

Not a consumer toy, and not “ChatGPT for kids”.

A product of TamerinTECH

Machine Learns — You Grow

ContextFriend is entering school pilots. We build it, we install it, and we hand the school the keys to its own brain.

tamerin.tech

Questions school operators actually ask

Does audio leave the school?

No. Audio is streamed on the school LAN and never sent to a cloud speech-to-text API. During a supervised pilot an on-premises archive can be enabled for quality analysis, and it is purged afterwards.

Does it work offline?

Teaching does. Updates, model downloads, remote administration and the optional cloud text-to-speech need a network path; the lesson itself does not.

Do you identify students?

No. Speaker labels are anonymous (Teacher, Speaker-2). There is no face recognition, no attendance tracking and no grading from audio.

Where does the data live?

On the Central Brain on the school LAN, or another local site the operator controls. Remote access to the portal goes through the school's VPN or tunnel — it is not a public SaaS URL.

What hardware do we need?

One strong local machine plus a Room Brain per room. A Mac mini in the 64 GB class is our current pilot profile, not a requirement — a similar on-premises box is fine. The Room Brain is Raspberry-Pi class with a speakerphone.

Is there a price list?

No public prices. We start with a pilot in your school, and the pilot tells us both what the right shape is.

How does this sit with GDPR and FERPA?

The architecture is on-premises: TamerinTECH does not host classroom data, so the school stays the controller for it. We do not claim a compliance badge — we give you an architecture your data protection officer can inspect.

Can the works council or the DSB inspect it?

Yes. There is a local admin overview of what is stored, what the retention is, and a purge button. Nothing about it requires our involvement.

Can rooms share memory?

Only if an operator explicitly links rooms or assigns shared collections — a workshop plenum, for example. The default is isolation.

Ask us something

Short question? Send it here.

TamerinTECH GmbH processes this form to answer your pilot request (double opt-in). Confirmed leads are kept up to 24 months. Unconfirmed submissions are deleted after 14 days. If you tick ads measurement, we send a hashed email and click ids to Google and/or Meta after you confirm — not required for a pilot. Details: Privacy. You can object at hello@context-friend.com. Erasure requests: we aim to complete them within 30 days. Privacy

Start with a pilot, not a purchase order.

Tell us about your school. We will tell you honestly whether ContextFriend fits it yet.

+49 8026 387 25 98