Open any file and read every byte of it, in hex and as text, patch the bytes in place, search for a signature, then save the file back out. Nothing is uploaded.
How it works
Choose a file or drop one on the File card, up to 8 MB of any kind. It is read into the tab and never sent anywhere, and Download writes the same bytes back out with your edits and nothing re-encoded.
Each row is an offset, 8, 16, 24 or 32 bytes in hex with a wider gap every eight, and one glyph per byte. The glyph column can be ASCII, Latin-1 or CP437, the DOS code page with its box-drawing characters; a byte with no glyph is a full stop. It is never UTF-8, because a character spread over several bytes has no single column to stand in.
Looks like, on the File card, matches the first bytes against about fifty signatures, PNG, JPEG, Zip, PDF, ELF, SQLite and WebAssembly among them. After RIFF it also reads the four bytes at offset 8, which tell WebP from WAVE and AVI; after ftyp it reads the brand, which tells an iPhone's HEIC photo and an AVIF from an MP4 video; and CA FE BA BE is named as both formats that use it. A file with no signature whose first 4 KB is printable is called UTF-8 or single-byte text.
The Inspector reads the bytes at the caret as binary, signed and unsigned integers from 8 to 64 bits, 32- and 64-bit floats, a Unix time and a GUID, little-endian unless switched. Below that, a selection, or the next 64 bytes, is decoded as UTF-8 and UTF-16 LE and spelled in hex and Base64.
Typing overwrites the byte under the caret, in hex or, with Type text, as a character of the chosen encoding. The length only changes through Insert bytes and Delete on the Edit card. Find takes hex or text, counts matches up to 1,000 and wraps at either end.
Examples
The start of a PNG
0000 89 50 4e 47 0d 0a 1a 0a 00 00 00 0d 49 48 44 52 .PNG........IHDR
0010 00 00 00 01 00 00 00 01 08 06 00 00 00 1f 15 c4 ................Looks like says “PNG image”. With the caret on offset 0010, the first byte of the width, the Inspector's UInt32 reads 16777216 in little-endian and 1 in big-endian. PNG stores its numbers big-endian, so this picture is one pixel wide.
An accented letter in three encodings
The text café au lait saved as UTF-8 is 13 bytes, because é takes two. The same row in each glyph column:
63 61 66 c3 a9 20 61 75 20 6c 61 69 74
ASCII caf.. au lait
Latin-1 café au lait
CP437 caf├⌐ au laitSelecting the 13 bytes puts café au lait on the Inspector's UTF-8 line and Y2Fmw6kgYXUgbGFpdA== on its Base64 line.
Common problems
A number in the Inspector is absurdly large
It is being read the wrong way round. PNG, JPEG and network protocols store numbers big-endian, most significant byte first; Zip, BMP and WAV store them little-endian. The page starts in little-endian, so switch the Inspector to Big-endian when the format says so.
Accented letters show as two dots
UTF-8 writes every character above U+007F as two to four bytes, and the glyph column draws one byte at a time, so ASCII gives full stops and Latin-1 gives pairs such as é. Select the bytes and read the Inspector's UTF-8 line. Text search follows the glyph column too: é in Latin-1 finds the byte e9, not UTF-8's c3 a9, so search for the hex pair; the Unicode inspector gives the UTF-8 bytes of any character.
Go to offset lands in the wrong place
A number made only of digits is read as decimal, even though the offsets are shown in hex: 100 goes to byte 100, which is 0x64. Write 0x100 for byte 256. Anything with a hex letter in it, such as 1a4, is read as hex.
An edit cannot be undone
There is no undo, only Put it back, which throws away every edit and returns the file as it arrived. Typing never changes the file's length, so a wrong keystroke can be fixed by typing the old value over it.
Frequently asked questions
Can I insert or delete bytes?
Yes, from the Edit card: Insert bytes at the caret or at the end, Fill the selection with a repeated pattern, or Delete the selected bytes. Typing in the dump only ever overwrites, so no offset moves unless you ask for it.
What can I type into the search box?
In hex mode, pairs of digits spaced however you like, with or without 0x or \x in front and commas between, so ff d8 ff and 0xff, 0xd8, 0xff find the same run. In text mode, characters that the chosen glyph encoding has a byte for.
Does Looks like read the whole file?
No. It compares magic numbers at fixed offsets, almost all within the first sixteen bytes, and never parses the format. A .docx shows as a Zip archive because that is what its first bytes are, and a text file that happens to begin with BM is called a BMP image.