When ChatGPT Needs a Chalkboard

What it does ChatTutor is a web-based AI tutor that hands large language models a set of visual teaching tools—an interactive math canvas and mindmap generator—instead of limiting them to text replies. It is built around the idea that STEM subjects need diagrams, graphs, and spatial reasoning, so the AI manipulates GeoGebra boards and Mermaid charts as if it were actually standing at a blackboard. Users can interact with the generated visuals directly in the browser while chatting with the model.

The interesting bit Most chatbots describe a parabola; ChatTutor draws one live and lets you tweak it. The project treats the LLM as an agent that can select and manipulate external tools, turning a conversation into a dynamic lesson rather than a static transcript.

Key highlights

• Integrates GeoGebra for live math visualization and Mermaid for mindmaps
• Multi-provider LLM support via the AI SDK (OpenAI, Anthropic, DeepSeek)
• Vue frontend with an ElysiaJS backend; self-hostable with Docker or Node/Bun
• Hosted version available at chattutor.app, but requires bringing your own API key
• Image upload depends on configuring external object storage (OSS)

Caveats

• You must supply your own LLM API keys; there is no bundled model
• Image upload is unavailable unless you configure OSS endpoint and credentials
• Requires a PostgreSQL database for persistence

Verdict Worth a look if you are building AI tutoring tools or teaching STEM subjects where a whiteboard beats a paragraph. If you just need a conversational chatbot, the extra infrastructure is probably overkill.