HTML Entity Encoder/Decoder

How to use it

1. Paste your text. A character reference anywhere in it preselects decoding, raw markup preselects encoding, and the badge states which call was made — click it, or press the s key, to overrule.
2. For encoding, set the two options above the output: Scope decides which characters become references (Minimal for the five markup-significant ones, Non-ASCII for everything past U+007F, Everything for all but ASCII letters and digits), and Style decides their spelling (Named where a name exists, Decimal, or Hex). Your choices survive swaps and re-runs.
3. For decoding there is nothing to configure — every recognized spelling is resolved in one pass, and chips report whatever needed tolerance: legacy names without semicolons, numeric remaps, or references the standard simply does not define.
4. Take the result with Copy or the .txt download, or fold the current state into a shareable link. To look up one character instead of converting a block, filter the reference table below by its glyph, name, codepoint or description.

Turn markup-significant characters into HTML entities, or collapse entity soup back into readable text. Direction is picked from what you paste, the encoder writes named, decimal or hex references at the scope you choose, and the decoder resolves all three spellings — including the legacy semicolon-less names — while declaring every tolerance as a chip. A searchable reference table of the common entities sits right below, and nothing you paste leaves this tab.

How it works

HTML keeps a few characters for itself: an ampersand opens a reference, angle brackets open and close tags, quotes delimit attribute values. To show those characters as content, a document spells them as character references — an ampersand, a payload, usually a semicolon. The payload comes in three forms: a name drawn from the standard's fixed table (—), the decimal codepoint (—), or the hexadecimal codepoint (—). All three mean the same em dash, and the decoder on this page resolves all three in a single pass, counting what it changed.

The name table is not open-ended. The WHATWG HTML Living Standard froze it at 2,231 entries — 2,125 names terminated by a semicolon plus 106 legacy exceptions — and publishes it as a machine-readable file, entities.json, from which this page's own table is generated. The 106 exceptions are an inheritance from the SGML era, when browsers accepted &copy or &lt with no terminator; retiring them would have broken decades of pages, so the standard instead fixed the rule: only names on that legacy list may match without a semicolon, and the tokenizer consumes the longest name it can. That rule is why &notanentity; decodes to ¬anentity; — &not is a legitimate legacy match, and the rest is ordinary text. When it happens here, a chip names the semicolon-less match rather than leaving you to puzzle over a stray ¬.

Numeric references carry their own compatibility layer. Windows tooling spent years emitting “ and — for curly quotes and em dashes; those numbers sit in the C1 control block, 0x80 through 0x9F, which contains no printable characters at all. HTML5 resolves that block through the Windows-1252 mapping — € comes out as € — because that is what the authors meant. NULL, unpaired surrogate codepoints and values past U+10FFFF become U+FFFD, the replacement character, since no document may carry them. Every such remap is declared as a chip instead of being silently absorbed.

Encoding is a policy question — which characters, spelled how. Minimal scope converts exactly the five markup-significant characters and leaves prose readable; Non-ASCII additionally converts everything above U+007F, for channels that mangle bytes; Everything converts all but ASCII letters and digits, the maximal form that makes invisible characters auditable. Style then picks the spelling. Named prefers the standard's name where one exists, settling ties for a codepoint by taking the shortest, lowercase-leaning candidate; the many codepoints without any name fall back to hex. And an astral character is always one reference for its full codepoint — 😀 is 😀, never a pair of surrogate references, because surrogates are a UTF-16 storage detail with no standing in a document.

Examples

<a href="?a=1&b=2">

<a href="?a=1&b=2">

© 2026 — café

© 2026 — café

™

™

Every example on this page is executed against the same engine that powers the tool, on every build.

Common errors

Input is larger than the 4 MB limit for this tool. For bigger payloads use the CLI: python3 -c 'import html,sys; sys.stdout.write(html.unescape(sys.stdin.read()))' < file.html

Four megabytes of entities is a scraped site or a template dump, not a snippet, and a browser tab is the wrong place to churn through it. The page refuses the whole paste — converting a truncated slice would produce a plausible-looking wrong answer — and your input stays put. Python's html module applies the same WHATWG table on your own machine; html.escape covers the encode direction.

`&nosuchentity;` at position 1 is not a known reference — left unchanged

The name between & and ; is not one of the standard's 2,231 entries, so the decoder passes it through byte-for-byte and points at it — one warning per distinct name, up to twenty, with an overflow note past that. Check the spelling and the case: names are case-sensitive, and while © and © both exist, mixed-case variants of most names do not.

This looks double-encoded — decode twice?

The input is full of &-prefixed reference shapes — the residue of a string that was escaped, then escaped again by a second system. Accepting the banner runs one more decode pass on the current output, and the action is undoable, so peeling a layer too many is a one-keystroke recovery rather than a re-paste.

Keyboard shortcuts

⌘/Ctrl + Enter

Run the conversion now — and the required confirmation once an input crosses the 2 MB manual band.

⌘/Ctrl + S

Download the current output as a .txt file; the help overlay notes that this overrides the browser's own save shortcut.

⌘/Ctrl + C

With focus in the tool and nothing selected, copy the whole output and announce it; any real selection copies just itself, natively.

⌘/Ctrl + Shift + C

Copy the complete output regardless of selection state.

s

Swap encode and decode (inert while typing in a field, during IME composition, and when single-key shortcuts are switched off).

t

Advance to the next color theme, under the same single-key guards.

? or ⌘/Ctrl + /

Open the shortcut overlay, browser-conflict notes included — and the reference table's filter field is reachable by Tab from the table heading.

Esc

Dismiss the topmost overlay only; pasted work is never touched.