The Chart & Graph Maker is 100% client-side. Your labels and values are held as ordinary state in the tab and passed to a pure rendering engine that runs in JavaScript — it never leaves your device. The live preview is an inline SVG generated from your data; the PNG download is produced by drawing the same geometry onto an in-memory <canvas> and calling toBlob; the SVG download is that same markup as a Blob. All three are created and downloaded in-memory with no network request, there is no analytics or tracking, and the tool works offline once cached. Close the tab and the data and the chart are gone.

PNG is a raster (pixel) format: this tool exports it at 2× or 3× the canvas size so it looks sharp on high-DPI screens and in documents, but enlarging it well beyond that will look soft because the pixels are fixed. SVG is a vector format — it stores the shapes and text mathematically, so it stays razor-sharp at any size and can be re-edited in a vector editor, which makes it the better choice for print or further design work. Both are generated from the identical chart geometry, so they look the same; the only caveat is that SVG references fonts by name, so if it is opened on a system without the same web-safe font the text metrics may shift slightly.

Pie and doughnut charts encode each value as an angle proportional to its share of the total, which makes them good for showing composition — how a single quantity splits into a handful of parts — but people judge angles and areas poorly, so they become unreliable with more than about five or six slices. Bar charts encode values as lengths along a common baseline, which the eye reads far more accurately, so they handle many categories, negative values and multiple series (grouped or stacked) cleanly and are the better default for comparisons. Line and area charts are for showing a trend or change across an ordered axis such as time. In this tool, pie and doughnut use only the first value column because they represent one whole; bar, line and area use every series you add.

Rather than ending the value axis exactly at your largest data point — which would produce awkward tick labels like 87 or 4,913 — the engine uses a Wilkinson-style “nice number” algorithm. It rounds the data range to a clean magnitude, picks a tick step that is a 1, 2, or 5 times a power of ten, and then extends the axis floor and ceiling to whole multiples of that step, always keeping the zero baseline so bar lengths stay honest. That is why the top gridline may sit a little above your maximum (for example an axis to 100 when your biggest value is 90): the trade is a small amount of empty headroom for gridlines on round, readable numbers, and big values are further compacted as 1.5K or 2M.