MLJAR Studio – Local AI Data Analyst That Saves Notebooks

TL;DR

TL;DR: MLJAR Studio is a cross-platform desktop app that turns natural-language data questions into executable Python notebooks — combining conversational AI with AutoML while keeping every analysis reproducible.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: mljar.com — verified by direct fetch
• Source repository: github.com/mljar/mljar-supervised — verified README, MIT license
• License: MIT — verified via GitHub API license.spdx_id
• HN launch thread: news.ycombinator.com/item?id=47985077 — 73 points, author pplonski86, 2026-05-02
• Source last checked: 2026-07-17

What Is MLJAR Studio?

MLJAR Studio is a desktop application built around the open-source mljar-supervised AutoML package. The core idea: you talk to your data in plain English, the AI generates and executes Python code locally, and the entire conversation is saved as a reproducible .ipynb notebook.

Unlike AI chat interfaces that vanish when you close the tab, every step in MLJAR Studio is a cell you can inspect, edit, and re-run. It sits between Jupyter Notebook (fully manual, flexible) and fully cloud-based AI tools (ephemeral, no artifacts).

The product comes from pplonski86, who has maintained mljar-supervised (MIT, ~3,275 GitHub stars) for several years. MLJAR Studio is the desktop frontend built on top of that library.

Pricing: $199 one-time purchase with a 7-day free trial.

Setup Workflow

Step 1: Download and Install

Download the installer for your OS from mljar.com. MLJAR Studio runs on macOS, Windows, and Linux.

Step 2: Local AI Backend (Optional)

By default MLJAR Studio can use:

• Ollama (zero data egress — all inference stays local)
• OpenAI API key (bring your own)
• MLJAR AI add-on (managed service)

To use Ollama locally:

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model (e.g. Llama 3)
ollama pull llama3

# Ensure Ollama is running
ollama serve

MLJAR Studio will auto-detect a running Ollama instance.

Step 3: Connect Your Data

MLJAR Studio works with:

• Files: CSV, Excel, Stata, Parquet
• Databases: PostgreSQL, MySQL, SQL Server, Snowflake, Databricks, Supabase

Step 4: Start a Conversation

1. Launch MLJAR Studio
2. Select a data source (file or database)
3. Type your question in natural language, e.g. “Build a classification model to predict churn, and show me feature importance”
4. MLJAR Studio generates Python code, executes it locally, and writes the result to a .ipynb notebook

The built-in AutoML handles algorithm selection and hyperparameter tuning for tabular data (classification, regression, multiclass).

Deeper Analysis

Reproducible Notebooks

The key differentiator is the .ipynb output. Each AI-assisted step produces a notebook cell — not just a chat message. You can:

• Re-run the full analysis with fresh data
• Modify individual cells
• Export to Mercury for sharing as a web app (Mercury is also from the same MLJAR org)

AutoML Modes

mljar-supervised (the underlying library) has four modes:

| Mode | Use Case | |---|---| | Explain | Understanding data — SHAP, permutation importance, decision trees | | Perform | Production-ready ML pipelines | | Compete | Competition-grade ensembles and stacking | | Optuna | Maximum-tuned models when compute time is not limited |

Database Connectivity

Connecting to Snowflake, Databricks, or Supabase means you can query live warehouse data without extracting it first — useful for recurring reporting workflows.

Practical Evaluation Checklist

• Runs fully offline (Ollama backend)
• Produces .ipynb artifacts, not ephemeral chat messages
• Built-in AutoML with four modes
• Database connectors for live queries
• Uses standard Python libraries (pandas, matplotlib, scikit-learn)
• One-time pricing ($199) vs subscription competitors

Security Notes

• Ollama backend: All inference stays on your machine. No data leaves your network.
• Cloud AI options: If using OpenAI or MLJAR’s own AI add-on, data may be transmitted. Review MLJAR’s privacy policy for the managed service.
• Local code execution: MLJAR Studio runs Python code locally. As with any tool that executes AI-generated code, use a sandboxed environment for untrusted data.

FAQ

Q: Is the mljar-supervised Python package itself free? A: Yes. The underlying mljar-supervised package is open source (MIT). MLJAR Studio is the commercial desktop product built on top of it.

Q: What operating systems are supported? A: macOS, Windows, and Linux.

Q: How does it compare to Jupyter Notebook + AI assistants? A: Jupyter gives you full control but requires manual coding. AI assistants (Cursor, Copilot in Jupyter) produce ephemeral chat messages. MLJAR Studio auto-generates notebook cells from natural language and tracks the full analysis as a reproducible file.

Q: Does it require a powerful machine? A: It runs inference locally via Ollama. For large AutoML searches (Compete/Optuna modes), a machine with decent CPU/GPU helps. The demo videos on YouTube show it running on a MacBook.

Q: Can I share the generated notebooks? A: Yes — they are standard .ipynb files. You can also convert them to a web app using Mercury (from the same org) for non-technical stakeholders.

Conclusion

MLJAR Studio targets data scientists and analysts who want the speed of conversational AI but the reproducibility of manual notebooks. The $199 one-time price is reasonable compared to cloud AutoML subscriptions, and the Ollama integration keeps data local. If you regularly build classification or regression models on tabular data and want a traceable artifact after every AI-assisted session, this is worth trying.

The 7-day trial gives you enough time to run one real analysis end-to-end and decide if the notebook-as-output workflow fits your process.

Try MLJAR Studio | GitHub (mljar-supervised)