Unicode Inspector

Break text into its characters and see each one's codepoint, UTF-8 bytes, and general category, plus the character and byte counts, then open any character for its full record: name, script, block, plane, bidirectional class, combining class, East Asian width, and the UTF-16/UTF-32, HTML and Rust/JavaScript/CSS escape forms. Useful for spotting invisible or look-alike characters, against a copy of the Unicode 16.0.0 character database built into the page.

A worked example

Every character of Hi 🌟 café─, broken down by the same Unicode 16.0.0 catalog the interactive version uses.

Per-character breakdown
Codepoint	Character	UTF-8	General category
U+0048	H	48	Lu (Uppercase Letter)
U+0069	i	69	Ll (Lowercase Letter)
U+0020	U+0020	20	Zs (Space Separator)
U+1F31F	🌟	F0 9F 8C 9F	So (Other Symbol)
U+0020	U+0020	20	Zs (Space Separator)
U+0063	c	63	Ll (Lowercase Letter)
U+0061	a	61	Ll (Lowercase Letter)
U+0066	f	66	Ll (Lowercase Letter)
U+00E9	é	C3 A9	Ll (Lowercase Letter)
U+2500	─	E2 94 80	So (Other Symbol)

Every attribute of one character

The full per-character record for U+2500, the box-drawing character the command line uses as its example (unicode-cli U+2500).

Name: BOX DRAWINGS LIGHT HORIZONTAL
Codepoint: U+2500
General category: So (Other Symbol)
Script: Common
Block: Box Drawing
Plane: 0 (Basic Multilingual Plane)
Bidirectional class: ON (Other Neutral)
Combining class: 0
Mirrored: no
East Asian width: A (Ambiguous)
UTF-8: E2 94 80
UTF-16: 2500
UTF-32: 00002500
HTML entity: ─ · ─
Rust escape: \u{2500}
JavaScript escape: \u2500
CSS escape: \002500

Security inspector

A worked scan of a sample that carries a right-to-left override, a zero-width space, a Cyrillic look-alike and a decomposed accent. 5 findings, led by bidirectional control characters (trojan-source risk).

Security findings
Severity	Finding	Why it matters	Characters
High	Bidirectional control characters (trojan-source risk)	This text contains bidirectional formatting controls. They reorder how characters are displayed without changing the stored order, so what a reader sees can differ from what a program parses or a compiler builds. In source code or an identifier that gap is the trojan-source technique. The characters are real Unicode controls with legitimate uses in right-to-left text; flagging them is not a claim of wrongdoing.	U+202E RIGHT-TO-LEFT OVERRIDE (right-to-left override); U+202C POP DIRECTIONAL FORMATTING (pop directional formatting)
Medium	Invisible or zero-width characters	These characters take little or no space and often render as nothing, so text can carry content a reader cannot see: hidden joins, a smuggled marker, or a space that is not the ordinary space. Each is shown here by its codepoint so it is never rendered as nothing.	U+200B ZERO WIDTH SPACE (zero-width space: takes no width and shows nothing)
Medium	Confusable characters (homoglyphs)	These characters look like ASCII letters or digits but are different codepoints, the technique behind look-alike domain names and spoofed identifiers. A name that reads as ASCII may not be. This uses the full Unicode confusables data, every script pair of it.	U+0430 CYRILLIC SMALL LETTER A (looks like ASCII "a")
High	Mixed scripts: Latin, Cyrillic	This text combines characters from more than one writing system. That is normal in many languages (Japanese mixes Han, Hiragana and Katakana; text often carries Latin punctuation), so a mix is not itself a problem. It matters when a word that looks like one script hides characters from another, the way a look-alike domain is built.	none
Informational	Not in NFC (composed) form	This text is not in NFC, Unicode's composed normalization form. It will look the same as its composed version but is a different sequence of codepoints, so a byte-for-byte comparison would call the two unequal. The normalization workbench shows all four forms side by side.	none

Look-alike characters

Unicode records which characters can be mistaken for which, across every writing system. pаypal.com and paypal.com reduce to the same comparison key (paypal.corn), which is why a look-alike domain works. 23 characters can be mistaken for the letter a alone.

Characters that look like the letter a
Codepoint	Character	Script	Name
U+0251	ɑ	Latin	LATIN SMALL LETTER ALPHA
U+03B1	α	Greek	GREEK SMALL LETTER ALPHA
U+0430	а	Cyrillic	CYRILLIC SMALL LETTER A
U+237A	⍺	Common	APL FUNCTIONAL SYMBOL ALPHA
U+FF41	ａ	Latin	FULLWIDTH LATIN SMALL LETTER A
U+1D41A	𝐚	Common	MATHEMATICAL BOLD SMALL A
U+1D44E	𝑎	Common	MATHEMATICAL ITALIC SMALL A
U+1D482	𝒂	Common	MATHEMATICAL BOLD ITALIC SMALL A
U+1D4B6	𝒶	Common	MATHEMATICAL SCRIPT SMALL A
U+1D4EA	𝓪	Common	MATHEMATICAL BOLD SCRIPT SMALL A
U+1D51E	𝔞	Common	MATHEMATICAL FRAKTUR SMALL A
U+1D552	𝕒	Common	MATHEMATICAL DOUBLE-STRUCK SMALL A
U+1D586	𝖆	Common	MATHEMATICAL BOLD FRAKTUR SMALL A
U+1D5BA	𝖺	Common	MATHEMATICAL SANS-SERIF SMALL A
U+1D5EE	𝗮	Common	MATHEMATICAL SANS-SERIF BOLD SMALL A
U+1D622	𝘢	Common	MATHEMATICAL SANS-SERIF ITALIC SMALL A
U+1D656	𝙖	Common	MATHEMATICAL SANS-SERIF BOLD ITALIC SMALL A
U+1D68A	𝚊	Common	MATHEMATICAL MONOSPACE SMALL A
U+1D6C2	𝛂	Common	MATHEMATICAL BOLD SMALL ALPHA
U+1D6FC	𝛼	Common	MATHEMATICAL ITALIC SMALL ALPHA
U+1D736	𝜶	Common	MATHEMATICAL BOLD ITALIC SMALL ALPHA
U+1D770	𝝰	Common	MATHEMATICAL SANS-SERIF BOLD SMALL ALPHA
U+1D7AA	𝞪	Common	MATHEMATICAL SANS-SERIF BOLD ITALIC SMALL ALPHA

Normalization workbench

The four normalization forms of ﬁlé ². They can render alike yet be different codepoint sequences, and compatibility forms fold distinctions away.

Normalization forms
Form	Text	Codepoints	Bytes	Scalars	Graphemes
NFC	ﬁlé ²	U+FB01 U+006C U+00E9 U+0020 U+00B2	9	5	5
NFD	ﬁlé ²	U+FB01 U+006C U+0065 U+0301 U+0020 U+00B2	10	6	5
NFKC	filé 2	U+0066 U+0069 U+006C U+00E9 U+0020 U+0032	7	6	6
NFKD	filé 2	U+0066 U+0069 U+006C U+0065 U+0301 U+0020 U+0032	8	7	6

NFC and NFD render the same but are different codepoint sequences: one composes marks into single characters, the other splits them out. A byte comparison would call them unequal.

Compatibility normalization (NFKC/NFKD) changes this text: it folds away distinctions like ligatures, super/subscripts, and fullwidth forms (fi becomes f i, squared becomes 2, fullwidth A becomes A). That is useful for loose matching, but it can collapse two strings a person reads as different into one, so it is the wrong choice for round-tripping exact text.

This is generated in browser and is not sent to pah.moi servers.

About this tool3 paragraphs

Past the per-character record there is a security pass: invisible and zero-width characters, always shown by codepoint rather than as nothing; bidirectional controls, with the trojan-source pattern named outright; mixed scripts; confusable homoglyphs; and text that is not in normal form. Its wording is deliberately educational rather than accusatory, because unusual text is usually just text.

A normalization workbench puts NFC, NFD, NFKC and NFKD side by side with the codepoint sequences before and after and the byte, scalar and grapheme counts, so a compatibility form that silently rewrites what you typed is visible instead of surprising. Search takes a codepoint, part of a name, or a descriptive word like arrow, box drawing or currency, and a terminal glyph matrix marks each cell's East Asian width so a box-drawing set can be picked without breaking a monospace layout.

The character database is a copy compiled into the page when the site was built, so this tool is exactly as current as that copy and never asks a server for anything. The confusable relation is the full UTS #39 table for single-codepoint sources, and a lookalike is a claim about shape, not intent: two characters can be confusable and both entirely legitimate. What you actually see also depends on the fonts your browser has, which is the one thing a codepoint cannot tell you.

Use it locally This tool has a native command line twin. Build unicode-cli from the site's source with:

cargo build --release --bin unicode-cli

Source and licence terms