Explore algebra, calculus, and physics through visual scenes you can question, manipulate, and learn from.
AlgeBench transforms AI from an answer machine into a learning companion. It builds interactive mathematical worlds, performs the tedious work, and reveals the structures worth exploring, while preserving the joy of discovery for the learner. Always available, it serves as an infinitely patient tutor, helping users develop genuine understanding rather than simply consume answers.
Spin up a lesson, move the sliders, rotate the model, listen to the explanation, then interrupt with what, how, or why. The narrator answers in real time while the visualization stays connected to the math.
Build intuition by manipulating equations, surfaces, vectors, and proofs instead of only memorizing formulas.
Create guided lessons that let students explore ideas at their own pace with narration, sliders, and visual checkpoints.
Teach yourself with an AI tutor that can explain the current scene, answer follow-up questions, and adapt to your level.
Prototype mathematical scenes, lesson formats, and AI-assisted learning workflows on top of an open source stack.
Spin, zoom, and inspect mathematical objects in scenes that respond to your input
Ask questions out loud while the tutor explains the scene and answers in context
Step through derivations with highlighted changes tied back to the visual model
See variables, operators, and relationships as an interactive map of the equation
Turn mathematical ideas into interactive lessons with narration, sliders, and visual structure
Study the code, extend the lesson format, or adapt the project for your own courses
Run AlgeBench on your own machine in one of two ways. Either way you'll need a free Gemini API key.
Requires Python 3.10+. The launcher sets up its virtual environment and installs dependencies on first run.
git clone https://github.com/ibenian/algebench
cd algebench
export GEMINI_API_KEY=your_key_here
./algebench
Then open localhost:8785 in your browser.
No clone, no Python setup — just Docker. Runs the prebuilt image from the Hugging Face Space registry.
docker run -it --pull=always -p 7860:7860 --platform=linux/amd64 \
-e GEMINI_API_KEY="your_key_here" \
registry.hf.space/ibenian-algebench:latest
Then open localhost:7860 in your browser. --pull=always fetches the newest build; --platform=linux/amd64 is needed on Apple Silicon and other ARM hosts.
GEMINI_API_KEY — billed to
your own account — instead of the shared public key on the hosted version, which is
rate-limited across all visitors.
GEMINI_API_KEY environment variable. You can create one for free at
Google AI Studio.
We recommend making a dedicated key for AlgeBench so you can track API usage separately.
See the Run Locally section above for the exact commands.
Watch walkthroughs, experiments, and new scene demos on the AlgeBench YouTube channel.
Try the hosted app instantly, or clone the repo to run it locally and contribute.