Jupyter AI

Jupyter AI is an extension for Jupyter Notebook and JupyterLab that brings AI capabilities directly into the notebook environment. It supports multiple LLMs and provides AI-powered code generation, explanation, and assistance within notebooks.

Yes, Jupyter AI is open source and free to use. You need to provide your own API keys for the LLM providers you want to use. There are no additional license costs.

Jupyter AI supports major LLM providers including OpenAI, Anthropic, Google, Cohere, and local models through various backends. You can switch between models using simple slash commands.

Yes, Jupyter AI can generate Python, R, Julia, and other language code directly in notebook cells. You describe what you need in natural language and the AI writes executable code.

Jupyter AI is open source and deeply integrated with the Jupyter ecosystem. Hex provides a more polished commercial experience with better collaboration. Jupyter AI suits open-source-oriented data scientists; Hex for data teams.

Yes, Jupyter AI can analyze error messages from failed code cells and explain what went wrong. It provides suggestions for fixing the error and can generate corrected code automatically.

Install the package with pip install jupyter-ai into the same environment as your JupyterLab, then enable the extension. After that, pick a provider such as OpenAI, Anthropic, or a local model through Ollama, and supply the matching API key in the chat panel or your environment variables. The %%ai magic command becomes available in cells once setup is complete, and the project documentation lists exact version requirements.

The %%ai magic command is the cell level interface: you prefix a cell with %%ai plus a model specification, write your instruction as plain text, and the cell output becomes the generated result, which can be code you then execute or text you keep as documentation. It complements the chat sidebar by making AI generation a repeatable, saved part of the notebook itself, so a data pipeline that needs a generated step keeps the prompt and the provenance in the notebook history where collaborators can see it.

Yes, in the typical setup. Jupyter AI is free open source software that acts as a layer connecting to AI providers, and you authenticate those providers under your own accounts and keys, whether that is a hosted model API or a local self hosted option. That structure means the software costs nothing but provider usage is billed by whichever provider you attach, so heavy notebook users should watch their API usage the same way they watch any other cloud spend. The flip side is full control over which provider and data boundary you use.

Yes, and it is one of the highest value everyday uses: because the assistant runs inside the notebook session, you ask for a chart against the dataframe you are actually working with, it generates the matplotlib or library specific plotting code in context, and you execute it directly in the next cell. Iteration is fast, asking for log scale, different colors, or a grouped variant is a follow up sentence rather than a documentation hunt. For analysts who know what they want to see but not the exact plotting syntax, this removes most of the friction.

The difference is context and proximity. A browser chatbot only knows what you paste into it; Jupyter AI runs inside the notebook session, so conversations happen next to the cells they concern and generated code lands where it can immediately run against your real variables and files. There is no copy-paste loop between two windows, and the notebook keeps the exchange in its history. For exploratory data work where every answer leads to another cell, staying inside the notebook is what makes the assistance feel continuous rather than intermittent.

Four cover most usage: generate boilerplate cells for loading and cleaning data so sessions start faster, explain a traceback inline instead of searching the error text, draft plotting code against the live dataframe, and produce first pass documentation or markdown summaries of what an analysis found. All of these share one pattern, the AI drafts and you verify, which keeps the notebook as the source of truth. Because it is free and open source, the realistic cost of adopting these workflows is just the provider usage behind your own keys.