Apr 17, 2026 · 8 min read · 🍎 Teachers How-To Guides

How to Set Up AI for Your School — Private, Free, and Safe for Students (2026)

Schools want AI. Parents want privacy. Cloud AI tools like ChatGPT send every prompt to external servers — including anything students type. For schools handling FERPA-protected data, that’s a compliance headache.

The solution: run AI on a school server. Student data never leaves the building. No subscriptions, no per-student fees, no data processing agreements with OpenAI.

This guide is for school IT administrators and tech-savvy educators who want to set up AI for their entire school.

Why Local AI for Schools

FERPA compliance — student data stays on school infrastructure
No per-student costs — one server, unlimited students
Content control — configure what the AI can and can’t discuss
No internet dependency — works even when the school WiFi is struggling
No accounts needed — students don’t need to sign up for anything
No age restrictions — ChatGPT requires users to be 13+. Local AI has no such restriction.
Full transparency — administrators can see every conversation

What You Need

Hardware Options

Setup	Cost	Supports	Best For
Repurposed desktop (16GB RAM)	$0 (existing hardware)	5-10 concurrent users	Single classroom
Mini PC (Intel NUC, 32GB RAM)	$400-600	10-20 concurrent users	Small school
Server with GPU (RTX 3090/4090)	$1,500-3,000	50+ concurrent users	Full school
Old laptop (8GB RAM)	$0	1-3 concurrent users	Teacher-only use

Start small: A repurposed desktop with 16GB RAM can serve a single classroom. If it works well, upgrade to a proper server for school-wide use.

Software (All Free)

Ollama — runs the AI model
Open WebUI — gives students a ChatGPT-like interface in their browser
Docker — runs Open WebUI in a container (optional but recommended)

Setup: Step by Step

Step 1: Install Ollama on the Server

# On the school server (Linux recommended)
curl -fsSL https://ollama.com/install.sh | sh

# Download a model
ollama pull llama3:8b

For a school with many concurrent users, use llama3:8b (fast, lower RAM) rather than larger models. Speed matters more than quality when 20 students are using it simultaneously.

If you have a GPU server, you can run qwen2.5:14b for better quality while maintaining speed.

Step 2: Install Open WebUI

Open WebUI gives students a familiar chat interface — it looks and feels like ChatGPT.

# Install Docker first (if not already installed)
# Ubuntu/Debian:
sudo apt install docker.io docker-compose -y

# Run Open WebUI
docker run -d \
  --name open-webui \
  --restart always \
  -p 3000:8080 \
  -v open-webui:/app/backend/data \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  -e WEBUI_AUTH=false \
  ghcr.io/open-webui/open-webui:main

Students access it at http://[server-ip]:3000 from any school device — Chromebooks, iPads, laptops, lab computers.

Note: WEBUI_AUTH=false means no login required. For school-wide deployment, you may want to enable auth and create class-based accounts. See Open WebUI docs for LDAP/Active Directory integration.

Step 3: Configure for Classroom Safety

This is the critical step. You need to control what the AI can discuss.

Set a system prompt in Open WebUI settings:

You are a helpful educational assistant for [school name]. 
Your role is to help students learn by explaining concepts clearly.

Rules you MUST follow:
- Never give direct answers to homework or test questions. Instead, 
  ask guiding questions that help the student figure it out themselves.
- Never discuss violence, weapons, drugs, alcohol, or sexual content.
- Never help students write essays or assignments for them. Help them 
  improve their OWN writing by giving feedback and suggestions.
- If a student asks something inappropriate, respond with: 
  "I can't help with that. Let's focus on your schoolwork. 
  What subject are you working on?"
- Never pretend to be a person, character, or authority figure.
- Always encourage students to verify information with their teacher 
  or textbook.
- If you're not sure about something, say so. Don't make things up.

You help with: math, science, English, history, geography, languages, 
and other school subjects. You explain concepts at a [grade level] level.

Customize per grade level: Create different system prompts for elementary, middle, and high school. Elementary prompts should use simpler language and be more restrictive. High school prompts can allow more nuanced discussion.

Step 4: Enable Conversation Logging

Teachers and administrators need to see what students are asking. In Open WebUI:

Go to Admin Settings → General
Enable conversation logging
Set up a review dashboard for teachers

This serves two purposes:

Safety monitoring — catch inappropriate use early
Learning insights — see what topics students are struggling with (if 15 students ask about fractions in one week, that’s useful data)

Classroom Use Cases

AI Tutor

Students ask the AI to explain concepts they’re struggling with:

Student: “I don’t understand photosynthesis”

AI (with system prompt): “Let’s break it down! Do you know what plants need to survive? Let’s start there and build up to the full process.”

The system prompt ensures the AI guides rather than lectures — it asks questions back, which is better pedagogy than just explaining.

Writing Feedback

Students paste their essays and get instant feedback:

Student prompt: “Can you give me feedback on this essay about the American Revolution?”

AI response: Feedback on structure, argument strength, grammar, and suggestions for improvement — without rewriting the essay for them (the system prompt prevents this).

Teacher benefit: Students get immediate feedback on drafts, which means they revise more before submitting. The teacher sees better final drafts.

Quiz Practice

Students ask the AI to quiz them before a test:

Student: “Quiz me on the water cycle for my science test tomorrow”

The AI generates questions, checks answers, and explains what the student got wrong. It’s like having a study buddy available 24/7 (well, during school hours if you restrict access).

Research Brainstorming

Students use the AI to brainstorm topics, create outlines, and organize research. The AI helps them structure their thinking without doing the work for them.

Language Practice

For ESL students or foreign language classes, the AI can hold conversations in the target language, correct grammar, and explain vocabulary — with infinite patience.

Security and Privacy Configuration

Network Setup

School Network
├── Student WiFi (VLAN 10)
│   └── Can access AI server on port 3000 only
├── Staff WiFi (VLAN 20)
│   └── Can access AI server admin panel
└── AI Server (VLAN 30)
    ├── No internet access (after initial setup)
    ├── Accepts connections on port 3000 from VLANs 10 and 20
    └── No outbound connections

Key points:

The AI server should NOT have internet access after the initial model download. This creates an air gap.
Only allow connections on port 3000 (the web UI). Block everything else.
Put the server on its own VLAN if possible.

Data Retention

Set conversation auto-delete after 30 days (or per your district’s data retention policy)
Don’t store student names in the system — use class codes or anonymous access
Back up conversation logs to a secure location for the retention period
Document your data handling in your school’s AI usage policy

Access Control

During school hours only — configure the server or firewall to only accept connections during school hours
On-campus only — don’t expose the server to the internet. It should only be accessible on the school network.
Teacher override — give teachers the ability to disable AI access for their classroom during tests

Addressing Parent and Administrator Concerns

“Is student data safe?” Yes. Nothing leaves the school network. The AI runs on a server in the building. No data goes to OpenAI, Google, or any third party. You can demonstrate this by showing the network configuration.

“Will students use it to cheat?” The system prompt is configured to guide learning, not give answers. Teachers can review all conversations. And the AI is a better cheating deterrent than a ban — students learn to use AI as a tool rather than a crutch.

“What about inappropriate content?” The system prompt blocks inappropriate topics. Conversation logging lets teachers monitor usage. And unlike ChatGPT, you control the guardrails completely.

“What does it cost?” After the initial hardware ($0 if repurposing existing equipment, $400-3,000 for new hardware), the ongoing cost is electricity only. Compare that to:

SchoolAI Pro: $6/teacher/month
ChatGPT Edu: varies by district
MagicSchool Pro: $9.99/teacher/month

For a school with 50 teachers, local AI saves $3,600-6,000/year in subscriptions.

Scaling Up

Single Classroom → Full School

Start with one classroom — one teacher, one class, one repurposed desktop
Gather feedback — what works, what doesn’t, what do teachers want
Expand to a grade level — upgrade to a mini PC or small server
Go school-wide — invest in a GPU server for 50+ concurrent users

Performance Tuning

If the AI is slow with many concurrent users:

Add a GPU — even a used RTX 3060 (12GB VRAM) dramatically improves speed
Use a smaller model — llama3:8b is faster than qwen2.5:14b and good enough for most student interactions
Limit concurrent users — Open WebUI can queue requests if the server is overloaded
Multiple model instances — on a powerful server, run multiple copies of the model for parallel processing

Troubleshooting

“Students say it’s too slow”

Check concurrent users: ollama ps shows active requests
Drop to a smaller model if needed
Add RAM or a GPU to the server
Set expectations: “It might take 10-15 seconds during busy times”

“The AI gave an inappropriate response”

Review the conversation log to understand what happened
Tighten the system prompt — add the specific topic to the block list
Consider adding a content filter layer (Open WebUI supports this)
Use it as a teaching moment about AI limitations

“Teachers don’t know how to use it”

Create a one-page guide with 5 specific use cases for their subject
Run a 30-minute workshop showing real examples
Start with the most tech-comfortable teachers and let them champion it
Share the conversation logs (anonymized) to show how students are actually using it

“IT department is concerned about security”

Show them the network diagram (no internet access, VLAN isolation)
Demonstrate that no data leaves the server
Provide the Docker configuration for their review
Offer to run it as a pilot with monitoring before full deployment

The Bottom Line

Local AI gives schools the benefits of AI in the classroom without the privacy risks of cloud services. Students get a ChatGPT-like experience, teachers get oversight and learning insights, and administrators get FERPA compliance and cost savings. The setup takes an afternoon and costs nothing beyond the hardware.

The schools that figure this out first will have a significant advantage — their students will learn to use AI as a thinking tool, not just a homework shortcut.

🛠️ Try our free teacher tools: Lesson Plan Generator · Quiz Generator · Report Card Comment Generator · Rubric Generator · IEP Goal Writer