notebook-assistant

Bring-your-own-LLM chat assistant for Jupyter notebooks. Load a DataFrame or markdown file, ask questions in plain English, get markdown-rendered answers right in your cell — with the provider you already pay for.

Supports OpenAI, Anthropic, Google Gemini, AWS Bedrock, and any OpenAI-compatible endpoint (e.g. Euron), routed through LiteLLM.

Install

pip install notebook-assistant

For AWS Bedrock also install the bedrock extra (adds boto3):

pip install "notebook-assistant[bedrock]"

Quickstart

Pick a provider, pass your key, ask a question.

OpenAI

from notebook_assistant import NotebookAssistant

assistant = NotebookAssistant(
    model="openai/gpt-4o",
    api_key="sk-...",   # or set OPENAI_API_KEY env var
)
assistant.ask("Give me three exploratory questions for a sales dataset.")

Anthropic

assistant = NotebookAssistant(
    model="anthropic/claude-3-5-sonnet-20241022",
    api_key="sk-ant-...",   # or ANTHROPIC_API_KEY
)

Google Gemini

assistant = NotebookAssistant(
    model="gemini/gemini-1.5-pro",
    api_key="...",   # or GEMINI_API_KEY
)

AWS Bedrock

assistant = NotebookAssistant(
    model="bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
    aws_region="ap-south-1",
    # aws_access_key_id / aws_secret_access_key optional — falls back to
    # boto3's standard credential chain (env vars, IAM role, ~/.aws/credentials).
)

You can also pass an inference-profile ARN as the model id.

Euron (OpenAI-compatible)

assistant = NotebookAssistant(
    model="openai/gpt-5.3-instant",
    api_key="...",   # your Euron API key
    api_base="https://api.euron.one/api/v1/euri",
)

Any other OpenAI-compatible endpoint works the same way — pass model="openai/<name>" with a matching api_base.

Loading context

import pandas as pd
from notebook_assistant import NotebookAssistant

df = pd.read_csv("sales.csv")

assistant = NotebookAssistant(model="openai/gpt-4o", api_key="sk-...")
assistant.add_context(
    instructions="Focus on month-over-month trends.",
    dataframes={"sales": df},
    markdown_path="docs/glossary.md",   # optional
)

assistant.ask("Which product saw the steepest decline last quarter?")

add_context accepts any combination of instructions, dataframes, and markdown_path. The content becomes the first user message in the conversation so every follow-up question sees it.

Interactive chat

assistant.chat()   # type 'exit' to quit

Conversation management

assistant.history   # list of {"role", "content"} dicts
assistant.reset()   # clear context and history

Configuration reference

Examples

• examples/quickstart.ipynb — end-to-end walkthrough with multiple providers.

Contributing

See CONTRIBUTING.md. Issues and PRs welcome.

License

MIT