jupyterpilot-ai 0.3.2

Agentic AI coding partner for IPython and Jupyter

🧠 JupyterPilot AI

JupyterPilot AI (jupyterpilot-ai) is a lightweight, provider-agnostic AI coding partner for IPython and Jupyter. It brings natural language code generation (%do) and automated error healing (%fix) to your notebook.

---

🚀 Features

• %do : Transform instructions into Python code instantly.
• %fix: Post-mortem self-healing for your last error.
• %review: Analyze the previous cell and get optimization suggestions or code reviews.
• Context Awareness: Remembers your last 3 cells for accurate variable usage.
• Data Observer (New): Automatically detects Pandas DataFrames, SQL engines, and MongoDB databases to provide schema context to the AI.
  • Smart MongoDB Introspection: Aggregates schema keys from the 20 most recent documents and extracts collection indexes, helping the AI write highly optimized queries.
• Hybrid Support: Switch between local Ollama and cloud GPT-4o/Claude via LiteLLM.

---

📥 Installation

pip install jupyterpilot-ai

Enable Data Observer (Optional)

To enable automatic schema detection for Pandas, SQL, and MongoDB, install the data extras:

pip install "jupyterpilot-ai[data]"

Enable the Extension

Add this to any cell in your notebook or your IPython startup script:

%load_ext jupyterpilot_ai

---

⚙️ Configuration

JupyterPilot AI looks for a configuration file at ~/.jupyterpilot/config.json.

Switching Modes

To switch between local and paid models, simply change the "mode" value in your config file:

{
    "mode": "cloud", 
    "local": {
        "url": "http://localhost:11434/api/generate",
        "model": "qwen2.5-coder:7b"
    },
    "cloud": {
        "provider": "openai",
        "model": "gpt-4o",
        "api_key": "sk-..."
    },
    "custom_link_ai": {
        "url": "",
        "model": "",
        "api_key": ""
    }
}

• Local Mode ("mode": "local"):

  • Uses Ollama by default.
  • Requires Ollama to be running (ollama run qwen2.5-coder:7b).
  • You can customize the url and model in the local block.

• Cloud Mode ("mode": "cloud"):

  • Uses LiteLLM to connect to paid providers.
  • Supports OpenAI, Anthropic (Claude), Google (Gemini), etc.
  • Specify your provider, model, and api_key.

• Custom Link AI Mode ("mode": "custom_link_ai"):

  • Connect to your custom AI API endpoint directly.
  • Set your custom url, model, and api_key.

---

🛠️ Usage

Basic Commands

• %do plot a sine wave
• %fix (after an error)
• %review (to review the last run cell)

Advanced Data Observer Example

If you have jupyterpilot-ai[data] installed, simply connecting to your database in a cell is enough for the AI to understand your schema.

# Cell 1: Connect to your database (Pandas, SQL, or Mongo)
from pymongo import MongoClient
client = MongoClient("mongodb://localhost:27017/")
db = client["Testing"]

# Cell 2: Ask the AI to write a query without explaining your collections or fields!
%do find the most recently active admin user

Behind the scenes, JupyterPilot automatically extracts the collection names, schema keys, and indices from db and injects them into the AI's context so it generates perfect code.

---

📬 Contact

Wajoud Noorani - [e-mail]

---

📄 License

MIT