Your private, local AI appliance
Turn a learning goal
into a lesson.
Keep AI local.
A teacher brings the learning goal. Your private AI appliance helps prepare the next lesson, explore a different approach, and keep useful work. The teacher reviews the result and decides what happens next.
Make the learning
goal visible.
3.NF.2 · Captured Kansas standardO1 · Locate one half by partitioning the whole interval into two equal parts.
Start with the teaching moment
What does your teacher
need to do next?
Choose a situation. See the teacher’s request, the work the appliance supports, and the judgment that stays with the educator.
“Help me teach one half using only the classroom board.”
Grade 3. Ten minutes. A clear learning goal, captured standard, and the materials actually available.
Illustrative scenario · prepared contentA structured first draft.
Use the selected standard and teacher context to request objectives, success criteria, a teaching sequence, and checks for understanding.
Check the mathematics, the objective-to-assessment connection, and whether the draft really honors the whiteboard-only constraint. Edit before using it.
A closer look
From a teaching need
to work you can use.
See the workflow behind the example: select the learning goal, generate a draft, review it, and save the work you choose.
A clear starting point.
Select Grade 3 and the captured Kansas 3.NF.2 standard, then describe the lesson you want to build.
- Goal: represent one half on a number line.
- Time and materials: 10 minutes; whiteboard and markers.
- Teacher instruction: students may point while seated.
The deployment model
A dedicated appliance.
AI that runs locally.
The teacher interface connects to an AI runtime on a dedicated local appliance. OpenWebUI manages the saved assistant, OpenVINO serves the language model, and PostgreSQL stores account-based lessons and versions.
Captured standards travel with the application. They remain identifiable source material, separate from the model’s instructional suggestions.
Explore the architecture ↗End-to-end disconnected operation, including sign-in, has not been qualification-tested. This product site is separate from the local appliance.
Measured lesson generation
Three teaching scenarios.
Twelve completed drafts.
12 serial lesson requests · 3 synthetic scenarios
Observed range: 35.6–49.8 sec.
Application adapter timing; excludes browser and saves.
4 requests per scenario · no errors or truncation
Real local model · one request at a time.
Teacher review remains essential.
Questions that matter.
Does AI Server Home replace instructional judgment?
No. Generated content is a draft for educator review. Teachers need to check accuracy, alignment, accessibility, and whether the response follows their instructions. Our recorded tests include material-constraint failures; a well-structured response is not automatically classroom-ready.
Where do the standards come from?
The application uses preserved Learning Commons captures of Kansas standards, with source records retained separately from AI suggestions. Mathematics, ELA, science, and social studies are represented. Framework versions and coverage have limits; the product does not claim a live lookup or universal standards coverage.
Can multiple teachers generate at once?
The current application adapter allows one active AI generation and rejects overlapping requests as busy. The first application tests document this behavior. Supported multi-user generation capacity has not been established.
Is this an Intel-certified product?
This site documents an Intel-based implementation and technical observations prepared for review. Intel program verification and approval are separate decisions. No certification, Intel endorsement, or incentive award is claimed.
See how it fits
Start with an educator’s workflow.
Walk through the product, discuss a deployment, and review the technical evidence with our team.