How to set up DeepTutor, the free AI tutor, and what it really costs to run

By M Farhan Malik - Opensrcs TeamGuide

Key takeaways

  • DeepTutor is a free, open source AI tutor. The app costs nothing; you only pay if you pick an online AI model.
  • On your own computer with Ollama it's completely free. A 16 GB laptop runs Qwen 3.5 9B well enough for everyday studying.
  • With an API key, normal studying costs about $1 to $3 a month on Claude Haiku 5.5 or GPT-6 Luna, and $15 to $60 on the top models.
  • Already pay for Claude or ChatGPT? DeepTutor can sign in with that plan instead of an API key.

DeepTutor is a free AI tutor that you install on your own computer. You give it your notes, PDFs and slides, and it answers questions from them, writes quizzes, solves problems step by step and plans what to study next. It was one of the fastest-growing projects on GitHub in November 2026, with more than 58,000 stars.

DeepTutor itself is free, with no paid plan. But it needs an AI model to think with, and that's where people get stuck. This guide covers installing it, the free route with a model on your own computer, which model your computer can actually run, and what the online models cost per month if you'd rather use those.

Our recommendation

Short on time? Pick one of these two. Both work well for everyday studying, and you can switch later in Settings.

Free on your own computer

Best forPrivate, and nothing to pay

Ollama with Qwen 3.5 9B for answers and nomic-embed-text to search your notes. Nothing leaves your computer. Good for chat, quizzes and study plans; slower than online models.

Model
qwen3.5:9b (6.6 GB download)
You need
16 GB of memory or an Apple Silicon Mac
Cost per month
Free
Online, pay as you go

Best forFast and smart on any computer

Claude Haiku 5.5 with an API key. Quick, good at step-by-step solving, and cheap. Switch to Claude Sonnet 5.5 only for the hardest problems.

Model
Claude Haiku 5.5
You need
An Anthropic API key and prepaid credit
Cost per month
About $1 to $3

Already pay for Claude Pro or ChatGPT Plus? Skip both and sign in with that plan instead (step 2).

Step 1: Install DeepTutor

Go to github.com/HKUDS/DeepTutor, open Releases on the right, and download the file for your computer:

  • Mac with Apple Silicon (M1 and newer): Open-DeepTutor-MacOS-Apple-Silicon.dmg
  • Older Intel Mac: Open-DeepTutor-MacOS-Intel.dmg
  • Windows: Open-DeepTutor-Windows-Setup.exe

On a Mac, open the file and drag DeepTutor into Applications. On Windows, run the setup file. The Windows app isn't signed yet, so SmartScreen may warn you the first time: click More info, then Run anyway. Only do this for the file you downloaded from the official HKUDS page.

The desktop app is marked "Preview", but it includes everything it needs, so you don't have to install Python or anything else.

Prefer your browser? If you use Docker, one command runs DeepTutor at http://127.0.0.1:3782:

Terminal
docker run --rm --name deeptutor -p 127.0.0.1:3782:3782 -v deeptutor-data:/app/data ghcr.io/hkuds/deeptutor:latest

Step 2: Choose where the AI comes from

DeepTutor gives you three ways to power it. You can switch between them any time.

  • Free, on your own computer. Run a model with Ollama or LM Studio. Nothing leaves your computer and nothing is billed, but it needs a reasonably modern computer, and it's slower and less clever than the best online models.
  • The plan you already pay for. If you have Claude Code signed in on your computer, DeepTutor can use your Claude Pro or Max plan. There's also a sign-in option for OpenAI Codex, which uses your ChatGPT account. No API key, and no extra bill.
  • An API key. Get a key from OpenAI, Anthropic, DeepSeek, Gemini or one of the dozens of other providers DeepTutor supports, and pay only for what you use.

Step 3a: The free route with Ollama

Ollama is a free app that downloads and runs AI models on your computer. Install it from ollama.com, then open a terminal (Terminal on a Mac, PowerShell on Windows) and download two models: one to talk to, and a small one that lets DeepTutor search your files.

Terminal
ollama pull qwen3.5:9b
ollama pull nomic-embed-text

Then connect them in DeepTutor:

  1. Open Settings, then Language models.
  2. Choose Ollama as the provider, leave the API key empty, and use the address http://localhost:11434/v1.
  3. Pick qwen3.5:9b as the model.
  4. Open Embedding models, choose Ollama, pick nomic-embed-text, and use http://localhost:11434/api/embed with dimension 768.

If you run DeepTutor in Docker, replace localhost with host.docker.internal in both addresses, because inside Docker "localhost" means the container, not your computer.

Which local model is enough?

It depends on your computer's memory, or on your graphics card if you have one. A model needs to fit in memory with room to spare, so here's what to pull:

  • 8 GB of memory, no graphics card: qwen3.5:4b (3.3 GB) or gemma4:e2b (4.6 GB). Fine for chat and simple quizzes, but slow, and it struggles with Research and Solve.
  • 16 GB of memory, an Apple Silicon Mac with 16 GB, or an 8 GB graphics card: qwen3.5:9b (6.6 GB). This is the best place to start for most people. gemma4:12b (7.7 GB) is a good alternative.
  • 32 GB or more on a Mac, or a 24 GB graphics card: qwen3.5:27b (17 GB) or gemma4:26b (16 GB). These get much closer to the online models for step-by-step solving and research.

For the embedding model, nomic-embed-text is small and works well for English notes. If your notes are in another language, try qwen3-embedding:0.6b or bge-m3 instead.

Pick your embedding model before you build a knowledge base. If you change it later, DeepTutor has to rebuild your knowledge bases from scratch.

Step 3b: The online route with an API key

Online models are faster, smarter and work on any computer, but you pay per use. You buy prepaid credit on the provider's website, create an API key there, and paste it into Settings, Language models in DeepTutor. These are the official addresses:

  • Claude (Anthropic): console.anthropic.com, address https://api.anthropic.com/v1
  • ChatGPT (OpenAI): platform.openai.com, address https://api.openai.com/v1
  • DeepSeek: platform.deepseek.com, address https://api.deepseek.com

Online models charge per "token", which is roughly three-quarters of a word. You pay for the tokens you send (your question, the recent chat and the parts of your notes DeepTutor attaches) and, at a higher price, for the tokens you get back.

What it costs per month

We worked out two kinds of month. These are estimates, because every question is different:

  • Light studying: about 20 questions a day. Each one sends around 8,000 tokens and gets about 1,000 back. That's roughly 5 million tokens in and 0.6 million out a month.
  • Regular studying: about 50 questions a day, plus two bigger Research or Solve tasks, which each use around 100,000 tokens. That's roughly 18 million tokens in and 2.4 million out a month.
Estimated monthly cost of running DeepTutor, in US dollarsLight studyingRegular studyingLower is better
NameLight studying and Regular studying
Claude Haiku 5.5
0.78, Light studying
3.00, Regular studying
GPT-6 Luna
0.78, Light studying
3.00, Regular studying
DeepSeek Flash
2.16, Light studying
8.28, Regular studying
GPT-6.1 Terra
7.80, Light studying
30.00, Regular studying
DeepSeek V4 Pro
8.71, Light studying
33.26, Regular studying
Claude Sonnet 5.5
15.60, Light studying
60.00, Regular studying
GPT-6.1 Sol
15.60, Light studying
60.00, Regular studying

Our estimate from official prices on 12 November 2026, per million tokens in and out: Haiku 5.5 and GPT-6 Luna $0.10/$0.50, DeepSeek Flash $0.30/$1.20 at peak, GPT-6.1 Terra $1/$5, DeepSeek V4 Pro $1.32/$3.96 at peak, Sonnet 5.5 and GPT-6.1 Sol $2/$10.

A few things bring these numbers down:

  • Off-peak DeepSeek is half price. DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays. Outside those hours, Flash would cost about $1 for a light month and $4 for a regular one.
  • Caching. When DeepTutor sends the same instructions and notes again, most providers charge a small fraction for the repeated part, so real bills are often lower than our estimate.
  • Mixing models. Use a cheap model for everyday chat and quizzes, and a top model only for hard problems.

Your embedding model costs almost nothing either way: free with Ollama, or a few cents a month online.

Which one should you pick?

  • Cheapest and good enough for most studying: Claude Haiku 5.5 or GPT-6 Luna. A regular month comes to about $3.
  • Best for hard maths, science and code: Claude Sonnet 5.5 or GPT-6.1 Sol. Use them when the cheap model gets stuck, or set a monthly spending limit on the provider's website.
  • Already paying for Claude or ChatGPT: sign in with that plan and pay nothing extra.
  • Want it free and private: Ollama with qwen3.5:9b, as in step 3a.

Step 4: Start studying

  1. Open Resources, then Knowledge Center, and press New knowledge base.
  2. Choose Create new, give it a name such as "Biology exam", and add your PDFs, slides, Word files or notes.
  3. Wait until the knowledge base shows as ready. Large PDFs can take a few minutes.
  4. Go back to Chat, pick your knowledge base, and ask a question about it. Then try Quiz for practice questions, or Mastery Path for a study plan.

Files you drop into a chat aren't added to a knowledge base automatically, so put anything you want to study again in the Knowledge Center.

Keep costs under control

  • Set a spending limit on your Anthropic, OpenAI or DeepSeek account before you start, so a long research task can't surprise you.
  • Start small. Add $5 of credit, use DeepTutor for a week, and check the usage page to see your real cost.
  • Keep your API key private. Never paste it into a chat or share a screenshot of your settings. If it leaks, delete it on the provider's website and make a new one.

Which setup is right for you?

  • A parent setting up a home tutor for your children: install DeepTutor and Ollama on the family computer with qwen3.5:9b. It's free, nothing your children type leaves your home, and there's no account or card to manage. Add their school notes and textbooks to a knowledge base so answers stay close to what they're learning. AI can still get things wrong and DeepTutor has no parental controls, so sit with younger children and check answers together.
  • A student on a laptop with 8 GB of memory: don't fight your laptop. Use Claude Haiku 5.5 or GPT-6 Luna for about $1 to $3 a month, or a small local model like qwen3.5:4b if it must be free.
  • A student with a newer laptop or Mac (16 GB or more): start free with Ollama and qwen3.5:9b. Add Claude Haiku 5.5 later if Research or Solve feels slow.
  • Hard maths, science or coding: use Claude Sonnet 5.5 or GPT-6.1 Sol for those sessions, and set a monthly spending limit.
  • Already paying for Claude or ChatGPT: sign in with that plan and pay nothing extra.
  • A teacher or study group sharing one tutor: run the Docker version on one computer, with a local model or one API key with a spending limit.

Frequently asked questions

Is DeepTutor free?

Yes. DeepTutor is free and open source under the Apache 2.0 licence, with no paid plan. It only costs money if you connect an online AI model, and then you pay that company, not DeepTutor.

Do I need my own API key?

Not always. You can run a free model on your own computer with Ollama or LM Studio, or sign in with a Claude or ChatGPT plan you already have. You only need an API key if you want to pay per use for an online model.

Can my laptop run DeepTutor for free?

The app runs on almost any computer. For a free local model, 16 GB of memory or an Apple Silicon Mac is the comfortable minimum with qwen3.5:9b. With 8 GB, a small model like qwen3.5:4b works for chat and quizzes but is slow.

Can my children use DeepTutor?

Yes, with you nearby. Set it up on the family computer with a free local model, and nothing they type leaves your home. DeepTutor has no parental controls, and any AI can give wrong answers, so check its answers with younger children.

Is DeepSeek the cheapest option?

Not any more. At the prices on 12 November 2026, Claude Haiku 5.5 and GPT-6 Luna are cheaper than DeepSeek Flash at peak times. Off-peak, DeepSeek Flash costs about the same.

  • DeepTutor

    58.5k GitHub stars

    A free AI tutor for your computer. It turns your notes and PDFs into quizzes, study plans and books, and can run on a free local AI model.

    • LLM & inference
    • RAG & agents
    • Apache 2.0
  • Ollama

    185k GitHub stars

    Run AI chat models like Gemma, Qwen and DeepSeek on your own computer with one command, for free and in private. Other apps can use them too.

    • LLM & inference
    • MIT License