Frequently Asked Questions

Everything you wanted to know about Muffin but were too busy converting CSVs to ask.

🧁

The short version

Muffin is a set of free, browser-based data tools for researchers, developers, and anyone who just needs to convert a file without signing up for yet another SaaS platform. Everything runs in your browser — your data never leaves your device. No exceptions.

// privacy & data
Does Muffin collect my data?

No. Not a single byte. All file processing happens entirely in your browser using JavaScript. When you upload a file or paste data, it is loaded into your browser's memory and processed locally — it is never sent to any server, database, or third-party service.

You can disconnect from the internet after the page loads and all tools will continue to work perfectly. That's how local it is.

Do you use cookies or track me?

We don't use any tracking cookies, analytics scripts, or fingerprinting. We don't know who you are, where you're from, or what you converted. We literally have no backend to log this information to.

The only external requests the page makes are to load fonts from Google Fonts and libraries from Cloudflare's CDN — standard stuff that every website uses, and neither of these can see your files.

Is it safe to process sensitive or confidential data here?

Yes — because nothing leaves your device. Your CSV, Excel, or notebook file is processed entirely in your browser's memory. Once you close the tab, it's gone. There's no cloud storage, no server logs, no database.

For the most sensitive data, you can even download Muffin and run it locally on your machine with no internet connection at all.

// data visualizer
What kind of CSV or Excel does the visualizer accept?

The visualizer accepts any standard flat CSV or Excel (.xlsx) file where the first row is a header row and each column has a name. It works best when your data is structured like this:

Good format (each column is one variable):
epoch, train_loss, val_loss, accuracy
1, 1.42, 1.51, 0.61
2, 1.18, 1.29, 0.68
...

It does not support multi-row headers (like merged cells in Excel where one header spans multiple columns). If your Excel file has that, flatten it to a single header row first.

Can I plot multiple models or columns at once?

Yes — this is fully supported. The Y Axis selector shows checkboxes for every column in your data. Just tick the ones you want to plot and they all appear as separate colored lines with a legend.

Example — comparing 5 models:
epoch, BERT_f1, RoBERTa_f1, DistilBERT_f1, XLNet_f1, AlBERT_f1
1, 0.59, 0.61, 0.56, 0.58, 0.55
...

Set X = epoch, tick all 5 model columns → you get 5 colored lines, each a different neon color, with a legend.

Each series automatically gets a distinct color: purple, green, pink, yellow, blue, orange, and more — always visible on the dark background.

What chart types are available?

Four chart types are supported:

  • Line chart — best for training curves, trends over time
  • Bar chart — best for comparing values across categories or models
  • Scatter plot — best for correlations between two variables
  • Histogram — best for seeing the distribution of a single column

You can also enable Highlight best value to mark the maximum or minimum point on any line with a star marker.

Can I download the chart?

Yes — click the PNG button above the chart to download it as a high-resolution 1200×700px PNG image, ready to drop into a paper, report, or presentation.

// notebook cleaner
How do I clean my Google Colab notebook?

In Google Colab: go to File → Download → Download .ipynb. Then upload that file to Muffin's Notebook Cleaner. It strips all outputs, resets execution counts, and removes Colab metadata — same result as running nbstripout, but no Python or terminal needed.

Which cleaning options should I use?

For most use cases — especially pushing to GitHub — check all three of these:

  • Remove cell outputs — clears all printed output, images, tables, error messages
  • Reset execution counts — resets the In[N] counters to null so diffs are clean
  • Strip notebook metadata — removes kernel info and Colab-specific metadata

If you're seeing an error like "the 'state' key is missing from metadata.widgets" on GitHub or nbviewer, also check Remove widget state — this removes the broken widget metadata that causes that error.

Can I clean multiple notebooks at once?

Yes. Click the Batch Clean tab, drop multiple .ipynb files at once, and all of them get cleaned with the same settings in one go. Each notebook gets its own download link showing how many outputs were removed and how much smaller the file got.

This is useful if you have a full folder of experiment notebooks you want to clean before pushing to GitHub.

Why should I clean notebooks before pushing to GitHub?

Three reasons: size (notebooks with outputs can be 100MB+), diffs (every run changes timestamps and execution counts making PRs unreadable), and privacy (outputs can accidentally contain API keys, file paths, or sensitive data you don't want public).

Cleaning before every commit is considered best practice. Tools like nbstripout do this automatically — Muffin's cleaner does the same thing on demand, in your browser.

// pdf merger
How do I merge multiple PDFs into one?

Go to Muffin's Image & PDF Converter and click the Merge PDFs tab. Drop your PDF files, reorder them with the ↑↓ arrows if needed, give your output a filename, and click Merge PDFs. Download button appears instantly.

Example — combining lecture slides:
You have Lecture_01.pdf, Lecture_02.pdf ... Lecture_10.pdf
Drop all 10 at once → they appear in numbered order → click Merge PDFs → download Combined_Lectures.pdf with all 10 lectures as one file, pages in order.
Does merging preserve the original PDF quality and formatting?

Yes, completely. Muffin uses PDF-lib to copy pages directly from each PDF into the merged file. The text, images, fonts, and layout are preserved exactly — nothing is re-rendered or compressed. The output is identical to the originals, just combined.

Is there a limit on how many PDFs I can merge?

No hard limit — you can merge as many PDFs as you want. The only practical limit is your browser's memory. Very large PDFs (100MB+ each) may be slow, but for typical lecture slides, research papers, or reports it will be fast.

The merging happens entirely in your browser, so no file size is "uploaded" anywhere.

Can I also convert images to PDF?

Yes — use the Images to PDF tab in the same tool. Drop multiple PNG, JPG, or WebP images and each one becomes a page in the PDF, in upload order. You can reorder with arrows before creating. Choose A4, Letter, or fit-to-image page size.

// data converter
What file formats does the converter support?

The Data Converter supports: CSV, JSON, JSONL (newline-delimited JSON), and Excel (.xlsx). You can convert between any combination — CSV → JSON, JSON → CSV, Excel → CSV, CSV → JSONL, JSONL → JSON, Excel → JSON, and so on.

// support & pricing
Is Muffin really free? What's the catch?

Really, genuinely free. No freemium tier, no file size limit behind a paywall, no "export to CSV requires Pro". The tools are just HTML, CSS, and JavaScript — they cost virtually nothing to host.

We're two people who built this in our spare time because we needed these tools ourselves. No VC funding, no team of 40. Just two people who got tired of sketchy upload-your-file websites.

How can I support Muffin?

The best ways to support Muffin:

  • ☕ Sponsor our coffee — If Muffin saved you an hour of work, consider buying us a coffee. (Ko-fi link coming soon!)
  • 🔗 Connect on LinkedIn — Say hi to Sneha Chakraborty or Divyansh Pathak. It genuinely makes our day.
  • 📢 Share it — Tell a colleague, post it in your team Slack, mention it in a paper footnote. Word of mouth is how small tools grow.
  • 🐛 Report bugs — If something breaks, reach out on LinkedIn. We actually fix things.
Can I sponsor Muffin or place an advertisement?

We're open to it! Muffin is built for data scientists and researchers — if your product is relevant to that audience, reach out on LinkedIn. We'd rather work with sponsors whose tools we'd actually use ourselves.

We'll never show intrusive ads or partner with anything spammy. Any sponsorship will be tasteful and clearly labelled.

I found a bug or have a feature request. How do I reach you?

Send a message on LinkedIn — Sneha Chakraborty or Divyansh Pathak. We read every message and take bug reports seriously. Feature requests that come up repeatedly actually get built.