Jupyter AI Features - Amazon SageMaker
From the chat UIFrom notebook cells
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Jupyter AI Features

You can access Jupyter AI capabilities through two distinct methods: using the chat UI or using magic commands within notebooks.

From the chat user interface AI assistant

The chat interface connects you with Jupyternaut, a conversational agent that uses the language model of your choice.

After launching a JupyterLab application installed with Jupyter AI, you can access the chat interface by choosing the chat icon in the left navigation panel. First-time users are prompted to configure their model. See Configure your model provider in the chat UI for configuration instructions.

Using the chat UI, you can:

  • Answer questions: For instance, you can ask Jupyternaut to create a Python function that adds CSV files to an Amazon S3 bucket. Subsequently, you can refine your answer with a follow-up question, such as adding a parameter to the function to choose the path where the files are written.

  • Interact with files in JupyterLab: You can include a portion of your notebook in your prompt by selecting it. Then, you can either replace it with the model's suggested answer or manually copy the answer to your clipboard.

  • Generate entire notebooks from prompts: By starting your prompt with /generate, you trigger a notebook generation process in the background without interrupting your use of Jupyternaut. A message containing the link to the new file is displayed upon completion of the process.

  • Learn from and ask questions about local files: Using the /learn command, you can teach an embedding model of your choice about local files and then ask questions about those files using the /ask command. Jupyter AI stores the embedded content in a local FAISS vector database, then uses retrieval-augmented generation (RAG) to provide answers based on what it has learned. To erase all previously learned information from your embedding model, use /learn -d.

Note

Amazon Q developer doesn't have the capability to generate notebooks from scratch.

For a complete list of features and detailed instructions on their usage, see the Jupyter AI chat interface documentation. To learn about how to configure access to a model in Jupyternaut, see Configure your model provider in the chat UI.

From notebook cells

Using %%ai and %ai magic commands, you can interact with the language model of your choice from your notebook cells or any IPython command line interface. The %%ai command applies your instructions to the entire cell, whereas %ai apply them to the specific line.

The following example illustrates an %%ai magic command invoking an Anthropic Claude model to output an HTML file containing the image of a white square with black borders.

%%ai anthropic:claude-v1.2 -f html
Create a square using SVG with a black border and white fill.

To learn about the syntax of each command, use %ai help. To list the providers and models supported by the extension, run %ai list.

For a complete list of features and detailed instructions on their usage, see the Jupyter AI magic commands documentation. In particular, you can customize the output format of your model using the -f or --format parameter, allow variable interpolation in prompts, including special In and Out variables, and more.

To learn about how to configure the access to a model, see Configure your model provider in a notebook.