The archive · Developer & Business Tools · Product decision · 2024–2026
Pretzel AI bets a Jupyter fork with AI, not a new tool, wins data work
Pretzel AI's browser data explorer drew 227 HN points in 2024; the two-person team's pivot forked Jupyter into an AI notebook, now 2.2k stars.
Pretzel AI
What the business is
Pretzel (Pretzel AI) makes Pretzel, an AGPL-licensed fork of Jupyter that adds AI code generation and editing, inline tab completion, sidebar chat and error fixing to notebooks; the two-person startup offers a free hosted version at pretzelai.app and plans a paid hosted version with company features.
How it started
On 2024-03-15 the project reached Hacker News as 'Show HN: Open-source, browser-local data exploration using DuckDB-WASM and PRQL' (227 points, 74 comments). The ambition was already to become the new default modern data tool, but talking to data professionals changed the route: 'we realized it will be very hard to get people to switch to a new tool, no matter how good. The best way to get people to switch is to not have them switch at all. That's why we decided to fork Jupyter — for the near zero switching costs.' The original in-browser data manipulation tool moved into the pretzelai_visual folder, and Pretzel became an AI-native Jupyter fork.
What happened
The README describes a two-person startup shipping quickly on top of the mature Jupyter codebase: cell-by-cell AI generation and editing with @variable context, inline tab completions (defaulting to Mistral's Codestral), a sidebar chat, an inject command, 'Fix Error with AI', and bring-your-own models from OpenAI, Anthropic, Ollama, Azure and Groq. The team chose AGPLv3 so a third party could not resell the code as SaaS without sharing changes, while keeping it free for individual and internal company use. Distribution pairs a free hosted server (accounts expire after 30 days of inactivity) with pip installation and Docker. Monetization, per the FAQ, is a hosted version with company-specific features such as access controls and data-source connectors; the individual version will not be made paid.
How it ended up
Still running as of the 2026-09-05 snapshot: the repository is live at 2.2k stars with 158 forks and 191 commits, the hosted app at pretzelai.app and product site at withpretzel.com are advertised, and no shutdown, acquisition or funding is disclosed in the reviewed material.
Background
Pretzel (Pretzel AI) is a two-person startup selling an AGPL-licensed fork of Jupyter that adds AI code generation and editing, inline tab completion, sidebar chat and error fixing to notebooks. Its bet, sharpened after the 2024-03-15 Show HN, was that the winning route to a new default data tool is a fork with near-zero switching costs rather than a brand-new tool — with a paid hosted enterprise version planned as the business.
The original product was browser-local data exploration built on DuckDB-WASM and PRQL; after talking to data professionals the team concluded people would not switch tools 'no matter how good'. Pretzel therefore became an improved Jupyter: same config, settings, keybindings and extensions, plus @-variable-aware cell editing, Codestral-backed inline completions, sidebar chat and 'Fix Error with AI', with users able to bring their own OpenAI, Anthropic, Ollama or Groq models.
The AGPL license was a deliberate business choice: the code stays free for individuals and internal company use, but anyone reselling it as SaaS must open-source their modifications — protecting the two-person startup's planned hosted product. Traction: the 2024-03-15 Show HN drew 227 points with 74 comments, and the 2026-09-05 GitHub snapshot shows 2.2k stars, 158 forks and 191 commits.
What has to be true
- Asking users to migrate tools fails no matter how good the new tool is; a fork inherits Jupyter's installed base and muscle memory.
- Shipping AI features where users already work (the notebook cell) shortens time-to-value to a single pip install.
- AGPL protects the planned hosted business from code-reselling while keeping the product free — a workable middle path for an open-source startup.
- Near-zero switching cost is a strong reason data professionals will at least try the fork, giving the two-person team a feedback loop to prioritize features.
What can be applied
If switching tools is the real barrier, don't build a better new tool — fork the incumbent, inherit its users, and add the missing layer (AI) where they already work.
Aftermath
As of the 2026-09-05 snapshot Pretzel is an active two-person open-source startup: the repository shows 2.2k stars, 158 forks and 191 commits, the hosted app at pretzelai.app and product site at withpretzel.com are live, and the FAQ still describes hosted enterprise features (access controls, connectors, collaboration, dashboards) as the monetization plan. No revenue, funding, customer-count or headcount figures appear in the reviewed material.
Sources
- Show HN: Open-source, browser-local data exploration using DuckDB-WASM and PRQL
- pretzelai/pretzelai — modern, open-source Jupyter alternative (2026-09-05 snapshot)
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card