Pretzel – AI-Powered Jupyter Fork

TL;DR

TL;DR: Pretzel is a drop-in Jupyter replacement that adds AI inline completions, a chat sidebar, and support for OpenAI, Anthropic, Ollama, and Groq — all in a single pip install.

What Is Pretzel?

Pretzel is a fork of Project Jupyter that layers AI-assisted coding directly into the notebook experience. It ships as a single Python package and opens a web interface via pretzel lab, maintaining full compatibility with existing Jupyter notebooks, extensions, and configuration.

The README describes it as “the modern replacement for Jupyter Notebooks” with AI code generation and editing built in.

Installation

Pretzel requires Python 3.9 or later.

pip install pretzelai
pretzel lab

This opens the Pretzel interface at http://localhost:8888/lab. All existing Jupyter config, settings, keybindings, and extensions carry over automatically.

For Windows users or environments where installation is problematic, a Docker option is available:

FROM python:3.9-slim
RUN apt-get update && apt-get install -y build-essential gcc
WORKDIR /root/pretzel
RUN pip install pretzelai
EXPOSE 8888
CMD ["pretzel", "lab", "--ip=0.0.0.0", "--allow-root", "--notebook-dir=/root/pretzel", "--ServerApp.allow_remote_access=True", "--ServerApp.token=''", "--no-browser"]

Build and run:

docker build -t pretzel .
docker run --name pretzel -p 8888:8888 pretzel

Access at http://localhost:8888/lab.

AI Features

Inline Completions

Start typing in any cell and Pretzel provides AI-powered tab completions. The system watches your code context and suggests completions inline without interrupting workflow.

Chat Sidebar

Press Cmd+K (Mac) or Ctrl+K (Linux/Windows) in any cell to open the AI chat interface. Alternatively, use Ctrl+Cmd+B (Mac) or Ctrl+Alt+B (Linux/Windows) to open the dedicated AI sidebar for code generation, debugging, and general questions.

Type @ to trigger autocomplete for function and variable names across your notebook.

Model Configuration

Pretzel works out-of-the-box with its default AI setup. To use your own models:

1. Open Settings → Pretzel AI Settings
2. Under AI Settings, pick models for chat and inline completion
3. The README recommends GPT-4 class models (GPT-4 Turbo, GPT-4o, Claude-3.5 Sonnet, Claude-3 Opus, Llama-3.1 405B) for best results
4. Under Configure AI Services, enter your API key or endpoint URL for each provider

Supported providers include OpenAI, Anthropic (Claude), Ollama (local models), and Groq. Requests go directly from your machine to the AI service — no data is sent to Pretzel’s servers.

Compatibility

Because Pretzel is a direct fork of Jupyter, it loads standard .ipynb notebooks and Jupyter extensions without modification. The claim is that “all of your Jupyter config, settings, keybindings, and extensions will work out of the box.”

The license is the Modified BSD License (same as Jupyter itself).

Project Status

• GitHub: github.com/pretzelai/pretzelai
• Stars: 2,187 (as of August 2026)
• Latest release: v4.2.11 (August 2026 — image support in sidebar and Cmd K)
• License: Modified BSD
• Repository pushed: December 2024

The roadmap includes native AI code generation similar to Cursor, frictionless realtime collaboration, SQL IDE support, visual analysis builder, Monaco-based code editing, and one-click dashboard sharing.

Source and Accuracy Notes

• Project page: pretzelai.org (main site) and pretzelai.app (hosted version)
• Source repository: github.com/pretzelai/pretzelai
• License: Modified BSD (verified from LICENSE file — same as Jupyter)
• HN launch thread: none verified for the main Pretzel launch. The closest is the July 2024 “Adding Mistral Codestral and GPT-4o to Jupyter Notebooks” post (269 pts) which was a plugin announcement rather than the product launch.
• Source last checked: 2026-08-14 (commit main)

FAQ

Q: Is my code sent to Pretzel’s servers? A: No. AI requests go directly from your machine to the provider (OpenAI, Anthropic, etc.). Pretzel states it does not store any code or data you send to the AI server.

Q: Can I use local models with Pretzel? A: Yes. Configure Ollama in Pretzel AI Settings to use local models without any API key.

Q: Do my existing Jupyter notebooks work with Pretzel? A: Yes. Pretzel is a drop-in replacement — it opens standard .ipynb files and respects Jupyter config and extensions.

Q: What Python version does Pretzel require? A: Python 3.9 or later.

Q: Is there a hosted version? A: Yes. pretzelai.app provides a free hosted option. No installation required.

Conclusion

Pretzel solves the “I want AI help in my notebook without leaving Jupyter” problem by forking Jupyter and embedding the AI layer directly. For data scientists already invested in the notebook paradigm, it removes the friction of switching to a separate AI coding tool. pip install pretzelai is the entire setup.

If you want to keep your existing notebooks and workflow but gain inline AI completions and a chat sidebar, Pretzel is the lowest-friction path — no migration, no new UI to learn, just open pretzel lab and code.