SADataCopilot

SADataCopilot is a VS Code extension for working with AI agents directly inside Jupyter notebooks.

It adds a Prompt cell type to ordinary .ipynb files. Prompt cells sit beside Python and Markdown cells, run like normal notebook cells, and can use the same live Python session as the rest of the notebook.

Objective

SADataCopilot keeps AI assistance inside the notebook workflow. Instead of copying notebook state into a separate chat, users can ask questions in place and let the agent inspect variables, run Python, and reason over results directly in the notebook.

Key Features

Prompt Cells

Use the + Prompt toolbar button to add an AI prompt to any Jupyter notebook. Ask questions, request analysis, generate code, or inspect notebook state without leaving the notebook.

Live Python Execution

Prompt cells can run Python through the same session used by notebook cells. Variables created earlier, such as df or model, are available to the agent while it responds.

Inline Results and Tool Transcript

Agent responses appear inline below the prompt cell, along with a readable transcript of tool calls, executed Python, and results.

Context Awareness

Users can include or exclude cells from the agent's context. Excluded cells still run normally, but the agent will not use them when answering future prompts.

Multiple Agent Backends

SADataCopilot supports multiple backend options:

Choose the backend from Settings -> SADataCopilot -> Backend or by updating copilotNotebook.backend.

Per-Cell Model Selection

Each prompt cell can use its own model setting. Available models depend on the selected backend:

• GitHub Copilot: GPT-5 mini and GPT-5.5 with configurable reasoning effort.
• Claude: Sonnet and Opus.
• Cursor: the configured Cursor agent model.

Notebook Context Header

The extension maintains a compact header at the top of each notebook showing token usage, estimated cost, prompt turns, and excluded cells.

Persistent History

For saved notebooks, prompt history and usage information are preserved across reloads.

Basic Workflow

1. Open a Jupyter notebook in VS Code.
2. Click + Prompt to add a prompt cell.
3. Ask a question or give the agent a task.
4. Run the prompt cell like any other notebook cell.
5. Review the inline answer, tool transcript, and updated context header.

Example prompts:

• "Summarize the dataframe and identify suspicious columns."
• "Check whether the train/test split has label leakage."
• "Plot the distribution of the target variable."
• "Write a Python cell that evaluates this model."