Introducing AI Tutor: Your Personal Learning Assistant Running Locally in Your Browser

What if you could have your own AI assistant that understands physics, engineering, computer science, math, health sciences, and more — running entirely on your device at zero cost? Today, we're making that a reality.

We're excited to announce the launch of AI Tutor — a feature that brings the power of large language models directly to your web browser. No cloud servers. No subscriptions. No data leaving your computer. Just you, your simulation, and an AI ready to help you understand the concepts behind what you're seeing.

---

Why We Built This

AI assistants like ChatGPT have transformed how people learn. But they come with trade-offs:

• Cost — Monthly subscriptions add up, especially for students
• Privacy — Your questions go to cloud servers you don't control
• Access — Requires constant internet connection
• Context — They don't know what you're looking at in your simulation

We wanted something different. An AI that:

• Runs completely locally on your device
• Costs nothing — no subscriptions, ever
• Knows the exact state of your simulation
• Works offline after the initial model download

So we built it.

---

How It Works

AI Tutor uses WebGPU, a new browser technology that lets websites access your GPU for high-performance computing. This means we can run sophisticated AI models directly in your browser — the same models that typically require expensive cloud servers.

Don't know what WebGPU is? That's totally fine — you don't need to. Just click the AI Tutor button and try downloading a model. If it works, great! If your browser shows an error, you might need to update to a newer version. The technical stuff happens behind the scenes; you just ask questions and get answers.

When you open any simulation on Simulations4All, you'll see the AI Tutor button in the corner. Click it, download your preferred model (just once — it stays cached), and start asking questions.

The AI sees your simulation's current state — every slider position, every parameter value. Ask "Why is the projectile going that high?" and it knows exactly which angle and velocity you're using.

What Are Open Source Models?

You might have heard of ChatGPT or Claude — those are "closed source" AI models owned by companies. You can only use them through their websites or apps, and you pay for access.

Open source models are different. They're AI models that researchers and companies have released for free, letting anyone download and use them. Think of it like the difference between renting a car (closed source) and owning one (open source). With open source, the model runs on your device, using your hardware.

The models we use — Qwen3 (from Alibaba) and Llama 3.2 (from Meta) — are among the best open source models available today. They're smaller than ChatGPT, so they can run on regular computers, but they're still remarkably capable for educational tasks like explaining physics concepts and solving problems.

---

Three Models for Every Device

Not everyone has a gaming PC with a powerful GPU. That's why we offer three model tiers, each optimized for different hardware:

Model	Size	Best For	Capabilities
Lite	~1.4 GB	Desktops, laptops, Chrome on iOS (experimental)	Quick answers, basic explanations
Standard	~2 GB	Laptops and desktops only	Detailed explanations, step-by-step solutions
Advanced	~2.3 GB	Desktops with modern GPUs (4GB+ VRAM)	Complex reasoning, multi-step physics problems

Our recommendation: Start with the Advanced model first. If your computer feels slow or unresponsive, unload it and try Standard. Still sluggish? Go with Lite. Everyone's hardware is different — you'll quickly find the right fit based on how your device responds.

Good news for older hardware: All three models can run on CPUs, not just GPUs. If your device doesn't have a dedicated graphics card (or has limited VRAM), the model automatically falls back to CPU processing. It's slower — responses might take 15-30 seconds instead of 5-10 — but it works. And you can unload the model anytime with one click in the AI Tutor interface if you need your computer's resources back.

---

What Each Model Can Do

Lite Model — Fast and Accessible

The Lite model is our smallest, designed to run on virtually any device with WebGPU support. It's perfect for:

• Quick concept checks: "What's the formula for kinetic energy?" or "What's the time complexity of bubble sort?"
• Simple calculations: "What's the velocity after 3 seconds?" or "Calculate the compound interest"
• Basic definitions: "What is Reynolds number?" or "Explain what a linked list is"

Example interaction:

You: What happens if I double the velocity?

Lite: Doubling velocity quadruples kinetic energy (KE = ½mv²). Range also increases by 4x for projectile motion since R ∝ v².

What to expect: Responses in 2-5 seconds. Good for straightforward questions where you need a quick answer.

Standard Model — Best Balance

The Standard model hits the sweet spot between speed and capability. It can:

• Explain concepts with context: "Why does the beam deflect more when I increase the length?" or "Why is merge sort faster than bubble sort?"
• Walk through calculations step-by-step
• Connect multiple concepts together across physics, engineering, computer science, and more
• Help debug your understanding: "I expected the range to be higher — why isn't it?" or "Why is my TDEE calculation so high?"

Example interaction:

You: Two projectiles are launched at 30° and 60° with the same speed. Which goes farther?

Standard: They travel the same distance! Using the range formula R = v²sin(2θ)/g:

• For 30°: sin(60°) = √3/2

• For 60°: sin(120°) = √3/2

Complementary angles (adding to 90°) always produce equal range with the same initial speed.

What to expect: Responses in 3-8 seconds. Handles most physics questions you'll encounter.

Advanced Model — Deep Reasoning

The Advanced model shines with complex, multi-step problems. It's ideal for:

• Working through exam-style problems in any subject
• Understanding non-intuitive behavior in physics, algorithms, or financial models
• Exploring "what if" scenarios in depth
• Connecting theory to real-world applications across all disciplines

Example interaction:

You: Why does the beam deflect proportionally to L³ when I change the length, but only linearly with force?

Advanced: Great observation! The beam deflection formula δ = FL³/(3EI) reveals why:

• Force (F) appears linearly because deflection is directly proportional to applied load — double the force, double the deflection.

• Length (L) appears cubed because bending involves both the moment (which increases with L) and the integration of curvature over the beam length (which adds another L² factor).

Think of it this way: a longer beam has more "leverage" working against it AND more material that can curve. These effects multiply, giving L³.

What to expect: Responses in 5-12 seconds. Worth the wait for complex questions.

---

Privacy First

Let's be clear about what "runs locally" means:

• Your questions never leave your device — all processing happens in your browser
• No accounts required — use AI Tutor without signing up (for a limited time!)
• No usage tracking — we can see which model you downloaded and that you interacted with AI Tutor, but nothing else. Your questions and conversations live entirely on your computer.
• Works offline — once downloaded, no internet needed

The model files are downloaded from our CDN (Cloudflare) once, then stored in your browser's cache. After that initial download, everything runs on your hardware.

---

What Hardware Do You Need?

Important: AI Tutor requires WebGPU, a modern browser technology for GPU computing. Not all devices support it yet. When you open AI Tutor, it automatically checks your device — if WebGPU isn't available, you'll see a clear message explaining that AI Tutor can't run on your current setup.

Here's what you need:

Requirement	Details
GPU	Any GPU with WebGPU support (most devices from 2020+)
GPU Memory	At least 2GB VRAM (4GB+ recommended for best experience)
Browser	Chrome 113+, Edge 113+, or Safari 18+ (macOS Sequoia)
Storage	~2.5GB for model files (downloaded once)
Mobile	Chrome on iOS: Lite model only (experimental). Safari on iOS: Not supported. Android: Lite model (experimental).

Mobile users (iOS): AI Tutor works best on desktops and laptops. On iPhones, it is experimental on Google Chrome (Lite model only) and is not supported on Safari. We are actively working on improving mobile support as WebGPU matures on iOS.

Will it work on my device? Just open any simulation and click the AI Tutor button. The tool automatically detects your GPU capabilities and tells you immediately whether AI Tutor is available. If your device supports WebGPU, you'll see which model is recommended for your hardware. If not, you'll see a clear explanation — no guessing required.

What if my device doesn't support WebGPU? You can still use all the other features of Simulations4All — the simulations themselves work on any modern browser. AI Tutor is an optional enhancement, not a requirement.

Apple Silicon note: M1/M2/M3 Macs work exceptionally well because they share memory between CPU and GPU, giving the AI more room to work.

Easy to control: Changed your mind? Need your computer's full power for something else? Just click the "Unload" button in the AI Tutor panel. The model releases all memory immediately. Load it again whenever you're ready — if it's cached, it starts instantly.

---

Always Improving

The AI landscape moves fast. New open-source models appear regularly, each better than the last. We commit to:

• Continuous updates — We'll swap in better models as they become available
• No disruption — Updates happen behind the scenes; your cached model keeps working
• Community input — Tell us which models you'd like to see supported

Our current models (Qwen3 and Llama 3.2) represent the best small-but-capable models available today. As the field advances, so will your AI Tutor.

---

Try It Now

Ready to meet your new study companion? Here's how to get started:

1. Open any simulation — try Projectile Motion for physics, Sorting Algorithms for computer science, or TDEE Calculator for health
2. Click the AI Tutor button in the bottom-right corner
3. Download the Advanced model first — see how your device handles it
4. If things feel slow, unload it and try Standard or Lite
5. Start asking questions!

Tips for best results:

• Be specific: "Why is the range 45 meters?" or "Explain the time complexity of this sort" works better than "explain this"
• Reference the simulation: "With these settings, what would happen if..."
• Ask follow-ups: The AI remembers your conversation context

Want the full walkthrough? Check out our Complete AI Tutor User Guide for detailed instructions on every feature, including memory settings, model management, and advanced controls.

---

Share Your Experience

We built this feature because we believe AI-powered learning should be accessible to everyone — not just those who can afford subscriptions.

If AI Tutor helps you understand physics better, share it with classmates. Show your teacher. Let others know that a free, private, powerful AI assistant exists right in their browser.

The future of education is local, private, and free. It starts today.

---

Disclaimer: AI Tutor is currently offered free of charge for a limited time. We reserve the right to introduce pricing or require account registration in the future. While we have designed AI Tutor to run safely within your browser, Simulations4All and its team are not responsible for any damages, data loss, system slowdowns, or other issues that may arise from using this feature. AI-generated responses are for educational purposes only and may contain errors — always verify important information with authoritative sources. By using AI Tutor, you acknowledge and accept these terms. That said, the models run entirely in your browser's sandbox environment, and we don't expect any issues — millions of users run these same open source models safely every day.