📓 notebook-crewai-agent

notebook-crewai-agent is an AI-powered EDA agent built with CrewAI, enhanced by fully-programmable Jupyter Notebook tools. This lets LLMs directly create, update, run, and control notebooks as part of an intelligent data analysis pipeline.

---

🚀 Key Features

• 🤖 LLM-driven autonomous agent that performs:

  • Data cleaning and missing value treatment
  • Feature engineering and selection
  • EDA (Exploratory Data Analysis) with visualizations
  • Final model suggestions

• 🧰 NotebookController class to programmatically control notebooks:

  • Create, insert, update, delete cells
  • Run specific cells or run all
  • Restart or start the kernel

• 🔌 Pluggable Notebook Tools (NOTEBOOK_TOOLS) can be independently integrated into any AI pipeline or toolchain (CrewAI, LangChain, etc.).

• 🌍 Supports all major LLMs – including OpenAI, Claude, Gemini, Ollama, and other LangChain-compatible models.

---

⚙️ Setup Instructions

1. Clone the repository

git clone https://github.com/S-k-Srivastava/notebook-crewai-agent.git
cd notebook-crewai-agent

2. Install dependencies using uv

Make sure uv is installed:

curl -LsSf https://astral.sh/uv/install.sh | sh

Then sync the environment:

uv sync

---

📚 Notebook Tools Overview

The core utility of this project is the NotebookController, a class that gives full control over Jupyter notebooks. It works both inside and outside CrewAI agents.

✅ Notebook File Operations

• create_notebook(path: str)
• load_notebook(path: str)
• save_notebook()
• export_to_format(format_type: str, output_path: Optional[str] = None)
• get_notebook_metadata()
• set_notebook_metadata(metadata: Dict)
• get_notebook_info()

---

🔧 Cell Management

• insert_cell(cell_type: str, source: str, index: Optional[int], cell_id: Optional[str])
• delete_cell(cell_id: str)
• delete_cell_by_index(index: int)
• get_cell(cell_id: str)
• get_cell_by_index(index: int)
• update_cell_source(cell_id: str, source: str)
• move_cell(cell_id: str, new_index: int)
• duplicate_cell(cell_id: str)
• clear_cell_output(cell_id: str)
• clear_all_outputs()
• set_cell_metadata(cell_id: str, metadata: Dict)

---

⚙️ Kernel Management

• start_kernel()
• stop_kernel()
• restart_kernel()
• interrupt_kernel()
• get_kernel_info()

---

🧠 Execution & Evaluation

• run_cell(cell_id: str, timeout: int = 30)
• run_cells(cell_ids: List[str], timeout: int = 30)
• run_all_cells(timeout: int = 30)
• run_cells_from_index(start_index: int, end_index: Optional[int], timeout: int = 30)

---

🗺️ Cell Info Utilities

• get_cell_count()
• get_cell_ids()
• get_code_cell_ids()
• get_cell_id_to_source_map()

---

🧼 Internal Helpers (Not intended for public use but useful internally)

• _update_cell_id_map()
• _generate_cell_id()
• _trigger_visual_update()

🧠 Agent Behavior Example

from crewai import Agent, Task, Crew
from modules.notebook_tools_crewai import NOTEBOOK_TOOLS

agent = Agent(
    role="Data Analyst and Scientist",
    goal="Complete EDA and dataset preparation",
    tools=NOTEBOOK_TOOLS,
    ...
)

The agent uses natural language tasks to:

• Access notebook tools
• Insert markdown/code cells
• Clean and transform data
• Visualize results
• Suggest ML models

---

🧪 Ideal Use Cases

• Jupyter notebook automation via AI
• AutoEDA and dataset preparation tools
• RAG or LangChain-based notebook agents
• LLM + notebook orchestration pipelines
• EDA-as-a-Service

---

📄 License

MIT © 2025 Saurav Srivastava