Help & Getting Started

What is pltchat?

pltchat turns a data file into a plot you refine by chatting. Upload a file and an LLM writes Python (matplotlib) code to plot it; every chat message tweaks the code. The code runs in your browser via Pyodide — your data is never executed on our servers, and only a small sample (column names, types, first rows) is sent to the LLM.

Quick start

Upload a file — CSV, TSV, TXT, JSON, XLSX or Parquet (max 50 MB). The first plot appears automatically after the Python environment loads (~a minute on first visit).
Refine by chat — describe changes in plain words: "make it a bar chart", "log scale on y", "only 2024 rows", "title: Revenue". Each message costs one token; automatic error fixes are free.
Edit the code directly — the generated Python is fully visible and editable. Press Run Code (or Ctrl+Enter) to execute your edits.
Zoom a plot — click any rendered plot to open it fullscreen; scroll to zoom, drag to pan, Esc to close.
Download — grab the plot as PNG/SVG/PDF or the code as a .py file to reproduce it anywhere.

Try an example dataset

No data file at hand? Download one of these, upload it, and start chatting:

monthly_revenue.csv — two years of seasonal sales (try: "revenue by month, color by region").
daily_weather.csv — a year of temperature and rain (try: "temperature over time with precipitation as bars").
plant_growth.csv — plant growth under two light regimes (try: "height vs day, one line per group").

History, versions, forks

Every change is saved as a version in the History panel. Click view to roll back to an earlier plot, or fork to branch off into a new project without losing the original.

Tokens

New accounts start with one free prompt. Each chat change costs one token; retries after an execution error are free. If you run out, use Buy tokens in the header. The balance is always shown next to your email.

Accounts & privacy

Login is passwordless: enter your email and we send a one-time code. Your files, prompts and generated code are stored until you delete them — the Delete account link in the footer removes everything (see the privacy policy for details).

Tips

Be specific: "scatter plot of price vs area, color by bedrooms" works better than "make it nicer".
If a plot fails, the auto-fixer tries twice for free; after that, read the error message — it usually says exactly what's wrong.
Big files are fine (50 MB), but plotting 10 million points will be slow — ask the chat to "downsample to 10k points" first.
Referring to what you see ("move the legend out of the upper left", "the two lines overlap")? Tick "Include the current plot" under the chat input — the rendered image is sent to the LLM with that prompt so it can see the plot. It's off by default because the image can show all your data points.
Turn off Auto-execute if you want to review code before it runs.

Something broken or missing? Use the Feedback button on the main page — it goes straight to the developers.