Show HN: Mljar Studio – local AI data analyst that saves analysis as notebooks

May 2, 2026•pplonski86•View Original

TL;DR Highlight

Ask in plain English; local Python runs and is saved as a reproducible notebook

Who Should Read

Data analysts and developers who can't send sensitive data to the cloud and want AI-assisted analysis locally

Core Mechanics

Plain-English questions become local Python runs
- You ask, the AI writes Python and runs it locally to show results.
- Every line of code is visible and editable, per the site.
Your data never leaves your machine
- The site claims 100% local execution with no external APIs required.
- It also says local LLMs are supported.
An AI agent runs ML experiments for you
- It improves notebooks step by step, testing ideas and searching for better models.
- It advertises model tuning, feature discovery, model comparison, and report generation.
Notebooks become web apps
- The open-source Mercury framework turns a notebook into a web app in one click.
- You can self-host it on your own server.
An AI sidebar lives inside the notebook
- It suggests Python code, data transformations, and visualizations.
- You decide what runs.

Community Reactions

Notebooks are bad for reproducibility, one commenter says
- Out-of-order execution and hidden state are the issues.
- They argue it fixes chat's non-reproducibility with an equally flawed format.
LLMs don't know kernel state when writing a cell, one opinion
- The problem isn't the format but out-of-order execution and silent kernel state.
- LLM generation makes it worse, they say.
Jupyter MCP Server gives a similar result, per one user
- Connected to Claude, it executes code on the live kernel.
- They have Claude write notebooks, run top to bottom, fix errors, and ping only when it works.
High-risk without a human in the loop, one comment argues
- Bad statistics can cost a lot; Zillow is cited.
- They worry data people shouldn't have to be expert code reviewers.
Commenters question differentiation
- One asks how it compares to open-source Deepnote.
- Others ask what the moat is versus one-shot Claude Code, and say real data work doesn't use notebooks.

How to Apply

Have Claude Code build a notebook
- Today, paste the prompt below with a sample CSV.
- Re-run the notebook top to bottom and confirm it passes.
Add a notebook reproducibility check to CI
- Add a step that restarts the kernel and runs all cells.
- Fix order-dependent cells when it fails.
Route sensitive data to a local-only path
- Analyze data that can't go to external APIs only with local tools.
- Have a human read and approve the generated code.

Try It

copy & run

# Paste into Claude Code or ChatGPT (simplified version of the Mljar Studio approach)

Build an analysis notebook (.ipynb) from this CSV: {{CSV path}}
Question: {{analysis question}}
Rules:
1. Write all analysis as Python code cells, each with a one-line description above it.
2. Cells must depend only on earlier cells in order. No hidden state.
3. When done, restart the kernel and run everything from the top to check for errors.
4. If an error occurs, fix it and re-run.
5. In the last cell, summarize conclusions and limitations in 3 lines.

Terminology

Jupyter notebookA file mixing code, outputs, and notes in one document, like a lab notebook.

AutoMLAutomation of model selection and tuning, like auto-tasting recipe variations.

MercuryMLJAR's open-source framework that turns notebooks into web apps.

Jupyter MCP ServerAn open-source server that lets an LLM run code on a live Jupyter kernel.

Related Resources

Show HN: Mljar Studio – local AI data analyst that saves analysis as notebooks

May 2, 2026•pplonski86•View Original

TL;DR Highlight

Ask in plain English; local Python runs and is saved as a reproducible notebook

Who Should Read

Data analysts and developers who can't send sensitive data to the cloud and want AI-assisted analysis locally

Core Mechanics

Plain-English questions become local Python runs
- You ask, the AI writes Python and runs it locally to show results.
- Every line of code is visible and editable, per the site.
Your data never leaves your machine
- The site claims 100% local execution with no external APIs required.
- It also says local LLMs are supported.
An AI agent runs ML experiments for you
- It improves notebooks step by step, testing ideas and searching for better models.
- It advertises model tuning, feature discovery, model comparison, and report generation.
Notebooks become web apps
- The open-source Mercury framework turns a notebook into a web app in one click.
- You can self-host it on your own server.
An AI sidebar lives inside the notebook
- It suggests Python code, data transformations, and visualizations.
- You decide what runs.

Community Reactions

Notebooks are bad for reproducibility, one commenter says
- Out-of-order execution and hidden state are the issues.
- They argue it fixes chat's non-reproducibility with an equally flawed format.
LLMs don't know kernel state when writing a cell, one opinion
- The problem isn't the format but out-of-order execution and silent kernel state.
- LLM generation makes it worse, they say.
Jupyter MCP Server gives a similar result, per one user
- Connected to Claude, it executes code on the live kernel.
- They have Claude write notebooks, run top to bottom, fix errors, and ping only when it works.
High-risk without a human in the loop, one comment argues
- Bad statistics can cost a lot; Zillow is cited.
- They worry data people shouldn't have to be expert code reviewers.
Commenters question differentiation
- One asks how it compares to open-source Deepnote.
- Others ask what the moat is versus one-shot Claude Code, and say real data work doesn't use notebooks.

How to Apply

Have Claude Code build a notebook
- Today, paste the prompt below with a sample CSV.
- Re-run the notebook top to bottom and confirm it passes.
Add a notebook reproducibility check to CI
- Add a step that restarts the kernel and runs all cells.
- Fix order-dependent cells when it fails.
Route sensitive data to a local-only path
- Analyze data that can't go to external APIs only with local tools.
- Have a human read and approve the generated code.

Try It

copy & run

# Paste into Claude Code or ChatGPT (simplified version of the Mljar Studio approach)

Build an analysis notebook (.ipynb) from this CSV: {{CSV path}}
Question: {{analysis question}}
Rules:
1. Write all analysis as Python code cells, each with a one-line description above it.
2. Cells must depend only on earlier cells in order. No hidden state.
3. When done, restart the kernel and run everything from the top to check for errors.
4. If an error occurs, fix it and re-run.
5. In the last cell, summarize conclusions and limitations in 3 lines.

Terminology

Jupyter notebookA file mixing code, outputs, and notes in one document, like a lab notebook.

AutoMLAutomation of model selection and tuning, like auto-tasting recipe variations.

MercuryMLJAR's open-source framework that turns notebooks into web apps.

Jupyter MCP ServerAn open-source server that lets an LLM run code on a live Jupyter kernel.

Show HN: Mljar Studio – local AI data analyst that saves analysis as notebooks

TL;DR Highlight

Who Should Read

Core Mechanics

Community Reactions

How to Apply

Try It

Terminology

Related Papers

Related Resources

Show HN: Mljar Studio – local AI data analyst that saves analysis as notebooks

TL;DR Highlight

Who Should Read

Core Mechanics

Community Reactions

How to Apply

Try It

Terminology

Related Papers

Related Resources