JSON viewer for files that choke other viewers

Use a viewer that draws only the branches you have opened, which is what this one does. A 5.00 MB document holds roughly 297,000 nodes; a viewer that mounts one element per node builds a third of a million DOM elements before you have read a single value, and that is the freeze. Here the first paint is the root plus its first 200 children — about 200 rows whatever the file weighs — and the pane never draws more than 3,000 rows at once. Parsing the same 5 MB takes under a fifth of a second, so the wait you notice is the file dialog, not the document.

Not to the standard: RFC 8259 defines an object as an unordered collection of name/value pairs, so two documents whose keys appear in a different sequence carry the same data. It matters to everything built on top of the standard, though — line-based diffs, checksums, cache keys and detached signatures all change when the sequence does. This viewer lists members in the order they were written rather than the order JavaScript would store them in, so what you read is the file, not a reshuffled copy of it.

Because the file really does contain it twice and hiding one would be a lie about your data. RFC 8259 only says object names SHOULD be unique, then warns that behaviour is unpredictable when they are not: JavaScript, Python, Go and jq all keep the last occurrence and silently drop the earlier ones, while some validators reject the document outright. Seeing both rows in the tree is usually the first clue that a template or a merge step is emitting a field twice.

There is no limit in the format itself — RFC 8259 lets every implementation set its own, and they differ wildly. This viewer reads 500 levels, .NET System.Text.Json defaults to 64, Go encoding/json stops at 10,000, and CPython json raises RecursionError somewhere under 1,000 because it recurses one stack frame at a time. If your document goes deeper than a handful of levels the breadcrumb above the tree is what keeps you oriented: it spells out the whole route from $ to the selected row and each step is clickable.

It is JSONPath, standardised as RFC 9535 in 2024, and it identifies the selected node inside the document. Identifier-shaped keys get dot form and everything else gets brackets, so a hyphenated header lands as $["content-type"] rather than an expression that would not parse. Drop the leading $ and it is a jq filter; keep the shape and it is close enough to paste into JavaScript or Python as an accessor chain.

Yes — it runs over the parsed document, not over the rows on screen, so a value buried in a folded branch is found and its ancestors are unfolded to show it. A complete pass over the 297,000 nodes of a 5 MB file takes about 15 milliseconds, and the tally stops counting at 500 hits so a one-letter query cannot stall the tab. Use the dropdown to limit the pass to keys or to values when the same word appears in both.

No. The file goes in through the browser's File API and is handled by code running inside this tab, so no part of it crosses the network — you can watch the network panel stay silent while a 5 MB document opens.