from2to
Conversion that fits in a pipe. Every command is named for what it does:
<from>2<to>.
$ hex2bin ff
11111111
$ nc example.com 4000 | dec2char
flag{d3c0d3d_1n_0n3_p1p3}
No Python REPL, no awk, no looking up printf format strings. One binary
installs ~83 short commands — hex2bin, dec2char, char2hex, hex2b64 —
each of which takes its input from the command line or from stdin.
And when you do not know what you are looking at yet, whatbase works it out:
$ whatbase -d bDNhcm5fdGgzX3IwcDM1
l3arn_th3_r0p35
Install
Every route gives you one binary, from2to; from2to install then creates the
per-conversion commands. See why the second step exists.
A prebuilt binary — no toolchain needed. Grab the archive for your platform
from the latest release,
check it against SHA256SUMS, and put from2to somewhere on your PATH. The
Linux builds are static musl, so they run in stripped-down containers and
rescue shells too.
Debian, Ubuntu, Kali, Parrot — .deb for amd64 and arm64:
$ sudo dpkg -i from2to_0.5.0_amd64.deb
$ from2to install
Fedora, RHEL — .rpm from the same release page, via sudo rpm -i.
Arch, BlackArch — from the AUR:
$ yay -S from2to-bin
$ from2to install
Windows — with Scoop:
> scoop bucket add from2to https://github.com/kayfgit/from2to
> scoop install from2to
> from2to install --dir "$env:USERPROFILE\scoop\shims"
The --dir matters here: Scoop puts the binary in its own app directory, which
is not the one on your PATH. from2to install tells you when the directory it
picked is not on PATH, so you will not be left guessing.
From source, with a Rust toolchain (1.85 or newer):
$ cargo install from2to
$ from2to install
The second command creates the per-conversion commands and prints where they went:
installed 84 commands via symlink in /home/you/.cargo/bin
Open a new shell so they are picked up, then check it worked:
$ hex2bin ff
11111111
Building from a clone works the same way — cargo install --path . from inside
the checkout, since that command reads Cargo.toml from the current directory.
Why the second step exists
The commands are links to the one executable, so the whole set costs no extra
disk space. On Unix they are symlinks; on Windows, hardlinks, because creating
a symlink there needs Developer Mode or an elevated prompt. --method symlink|hardlink|copy overrides that, and it falls back through the list if
your filesystem will not do the first choice.
It is a separate step because writing sixty-odd files into a directory on your
PATH should be something you ask for, not something a package manager does to
you behind your back.
Installing somewhere else
By default the commands go next to the binary. When that directory is not
yours to write to — which is the case for the .deb, .rpm and AUR packages,
where the binary lands in /usr/bin — they go to $XDG_BIN_HOME, or
~/.local/bin if that is unset. Either way the chosen directory is printed,
and you are told if it is not on your PATH.
from2to install --dir <path> puts the commands wherever you like — useful if
you would rather keep ~/.cargo/bin tidy. Putting that directory on your PATH
is then up to you. --dry-run lists what it would create without touching
anything.
Updating and removing
$ cargo install from2to --force
On Unix that is the whole update: the commands are symlinks, so they follow the
new binary and keep working. On Windows they are hardlinks pointing at the
old executable, so run from2to install again afterwards to refresh them — it
needs no flags.
from2to uninstall removes them again. Both directions identify their own work
by a marker compiled into the binary, so they recognize any version they
installed while leaving an unrelated hex2bin of your own strictly alone —
that one needs --force before install will touch it.
Commands
| Group | Names |
|---|---|
| Number bases | bin oct dec hex, in every pair: hex2bin, dec2hex, bin2oct, … |
| Arithmetic | any number base to itself: hex2hex '4438+1a', dec2dec, … |
| Text | char (alias ascii): dec2char, hex2char, char2hex, dec2ascii, … |
| Encodings | b64 b32: hex2b64, b642char, char2b32, b642hex, … |
| Escaped hex | esc (alias fhex): char2esc, esc2char, hex2esc, … |
| Generic | f2t <from> <to> for any base 2–36 |
| Byte order | le/be on either half: lehex2dec, hex2lebin, char2lehex, … |
| Detection | whatbase <text> — when you do not know the format yet |
| Shorthands | f2t 1628 = hex→oct, f2t 10216 = dec→hex, … (not installed as commands) |
from2to list prints all of them.
The numeric shorthands
<from>2<to>, using 2 as the word "to": 1628 is 16→8. These exist only for
bases 2, 8, 10 and 16, and that limit is deliberate. With 2 serving as
both a base and the separator, those sixteen names each have exactly one valid
reading — but 216 would be both 2→16 and 21→6. The restriction is what keeps
them unambiguous, so the set cannot be extended. Use f2t 21 6 for the rest.
They are not installed as commands. A file called 822 or 10216 sitting
in a directory on your PATH is a poor citizen, and it is the one name shape a
package manager would rightly object to. They are only a spelling f2t
understands:
$ f2t 1628 ff
377
That also settles an old wart: PowerShell parses a bare 1628 as the number
1628 rather than a command, so it could never have run them without ceremony.
f2t 1628 ff works everywhere.
Generic form
$ f2t 36 10 zz # any base 2-36
1295
$ f2t hex bin ff # word names work too
11111111
$ f2t hex2bin ff # so does a whole command name
11111111
When you don't know the format: whatbase
Every command above needs you to already know what you are holding. whatbase
is the other direction — paste the blob and be told what it plausibly is:
$ whatbase bDNhcm5fdGgzX3IwcDM1
b64 l3arn_th3_r0p35
also reads as b64url (identical here)
The answer, and what it says. A positional base that does not spell anything gives you the number instead, which is almost always what you wanted next:
$ whatbase deadbeef
hex 3735928559
A second line only appears when it is earned — so silence is itself a claim of confidence. Two things earn it: a twin that decodes identically, as above, or a rival close enough that choosing between them is a real decision:
$ whatbase 123456
dec 123456
also reads as oct (42798)
That is the case worth being loud about. 123456 is a perfectly good octal
number meaning something completely different, and the scores are 53 to 49 —
too close to quietly pick one. Meanwhile deadbeef scores hex 59 to char 35
and is not in any doubt, so nothing is added.
--why shows how it decided — the scores, the evidence, and the command that
finishes the job:
$ whatbase --why bDNhcm5fdGgzX3IwcDM1
b64 96% l3arn_th3_r0p35
all 20 characters are in the base64 alphabet
length 20 is a multiple of 4
decodes to printable text
b64url reads the same here — no + / - _ to tell them apart
try: b642char bDNhcm5fdGgzX3IwcDM1
could also be: char 26%
f2t detect is the same thing spelled out, for when you have not installed the
shims.
| Flag | Effect |
|---|---|
-w, --why |
show the scores, the evidence and the command to run |
-d, --decode |
skip the report, print what the best guess decodes to |
-q, --quiet |
print just the format name, for scripts |
-a, --all |
every candidate, including the implausible ones |
--to <fmt> |
with --decode, decode into something other than char |
So the one-liner for a challenge string is:
$ whatbase -d bDNhcm5fdGgzX3IwcDM1
l3arn_th3_r0p35
and the name it prints is one the converter accepts, so the two compose:
$ f2t "$(whatbase -q "$blob")" char "$blob"
How it decides
Three questions, none of which is enough alone:
- Does the alphabet admit it? A hard gate — one
zrules out hex. - Is it the narrowest alphabet that admits it?
01010101sits inside base64's alphabet perfectly happily, but nobody writes base64 that way. This is what makes the answerbin, and what stops1234567890reading as hex. - Does the payload look like anything? The question that separates base64 from base32hex on a string both of them accept: only one decodes to words.
Question 3 counts asymmetrically, on purpose. For base64 and base32, a decode
that produces noise is real evidence against the guess — those encodings exist
to carry something. For a positional base it is not: 1234567890 is a perfectly
good decimal number and owes nobody a printable decoding.
The score is a confidence, and only means anything next to the other
candidates — which is exactly why it is not in the default output. 59% on its
own tells you nothing; 59 against 25 tells you the answer is not in doubt.
--why shows the comparison, and -a adds every reading including the
implausible ones:
$ whatbase --why -a ff
hex 59%
1 token of nothing but base-16 digits
2 digits — a whole number of bytes
the decoding is not obviously text
try: hex2char ff
char 25%
b32 23%
b64 20%
b32hex 20%
-a on its own gives the same list, terse — one line per reading:
$ whatbase -a 123456
dec 123456
oct 42798
hex 1193046
char (already text)
b64
Two deliberate limits. whatbase never guesses at ciphers — a rot13 string is
text, and it will tell you it is text. And an interior space rules out the
encodings entirely, since it is the strongest hint in the string that you are
looking at words; line breaks do not, so wrapped base64 still works.
Input
Values come from the arguments, or from stdin when there are none:
$ hex2dec ff 10
255
16
$ printf '255\n16\n' | hex2dec
597
22
Numbers may be separated by whitespace, newlines or commas, and may carry a
0x / 0b / 0o prefix or _ spacers. --strict turns that leniency off.
For char input, whitespace is data, not a separator — except for one
trailing newline from a pipe, which is dropped so that echo hi | char2hex
does the obvious thing.
Output
Output streams as it is produced, so a slow producer shows up live:
$ nc example.com 4000 | dec2char # prints as bytes arrive, not at EOF
The one exception is base64/base32 output, which has to see the whole input before it can place its padding.
Separators are chosen to be the useful thing rather than a uniform thing:
| Conversion | Default | Why |
|---|---|---|
hex2dec |
newline per result | one line in, one line out |
dec2char |
nothing | the point is to reassemble a flag |
char2hex |
nothing | 68656c6c6f reads like xxd -p |
char2dec |
space | 104 101 108 108 111 is unreadable otherwise |
-d <s> sets it explicitly, -j joins with nothing.
Leading zeros
Between two power-of-two bases the digits you wrote carry real width information, so it is preserved:
$ hex2bin 0f
00001111
$ hex2bin -W 0f # bare numeric
1111
$ hex2bin -w 12 0f
000000001111
The same rule keeps a leading zero byte alive through an encoding:
$ hex2b64 0068656c6c6f
AGhlbGxv
$ hex2b64 68656c6c6f
aGVsbG8=
Base 10 has no such structure, so decimal results are always bare.
Arithmetic
Reading disassembly means doing math on addresses, and that used to be two conversions wrapped around a step you did in your head. Now it is one command, because a value can be an expression:
$ hex2hex '4438+1a'
4452
$ hex2hex '(4500 - 4438)/2' # how many words is that jump?
64
$ dec2hex '1000*4'
fa0
Bare literals are read in the source base, and the answer comes out in the target one — so the whole thing stays in hex, or crosses over, exactly as the command name says:
$ hex2dec '4438+1a'
17490
$ hex2bin 'ff & 0f'
1111
Quote your expressions. Almost every operator — * & | < > ( )
— is also a shell metacharacter, and an unquoted hex2hex '1000*4' is a glob
error rather than a number. A leading unary minus needs --:
$ hex2hex -- '-5+3'
-2
Subtraction is spelled with a space. A - with digits hard up against it
on both sides is punctuation inside one value, not an operator:
$ hex2hex '4500 - 4438'
c8
$ hex2hex '4500-4438'
from2to: '4500-4438': a '-' between two digits is part of the value, not
subtraction (a UUID or a MAC address is written that way, and would otherwise
read as a sum); put spaces around it to subtract
The rule exists because hex-shaped identifiers are punctuated with dashes and every other character in them is a hex digit, so the dash is the only thing that tells them apart from a chain of subtractions:
$ hex2dec 550e8400-e29b-41d4-a716-446655440000 # a UUID, not a sum
$ hex2dec 00-11-22-33-44-55 # a MAC address, not a sum
Both are an error, as they were before arithmetic existed. Answering them with
a plausible number and an exit status of zero is the one thing this tool will
not do. Only - needs the rule — nothing is written with a + in the middle
of it, so 4438+1a needs no spaces.
The operators are C's, and so is their precedence — loosest first:
| Operators | |
|---|---|
| loosest | | |
^ (exclusive or, not a power) |
|
& |
|
<< >> |
|
+ - |
|
* / % |
|
| tightest | unary - ~, then ( ) |
Shifts binding looser than + is a famous C trap — 1<<2+3 is 32, not 7 — and
it is reproduced deliberately. The point of this is to agree with the code you
are reading, not to have better opinions than it.
Division truncates toward zero, so -7/2 is -3, which is what the debugger
next to you says.
A prefix never overrides the source base. 0b1 in hex is 0xb1 — b is a hex
digit — and it means the same thing on its own as it does inside an expression:
$ hex2dec 0b1
177
$ hex2dec '0b1+0'
177
That is the whole rule, and it is worth stating plainly because the alternative
is worse. In hex, 0d26 is already the perfectly good number 0x0d26; letting
a prefix win would mean the tool silently choosing one of two readings of your
address. To add decimal 26 to a hex address, write 1a.
Negatives, and --word as a register
Without --word the arithmetic is signed and arbitrarily large, so a negative
result simply prints with a sign:
$ hex2hex '4438-4500'
-c8
$ hex2dec '1-5'
-4
With --word N the whole thing happens in an unsigned N-byte register
instead, wrapping after every operation — which is what the 16-bit machine you
are looking at actually does:
$ hex2hex --word 2 '4438-4500'
ff38
$ hex2hex --word 2 '~0'
ffff
$ dec2hex --word 4 '0-1'
ffffffff
In a register >> is a logical shift and -x is the two's complement; without
one, >> sign-extends and ~x is -x-1. The register width is also the
written width, so the result is padded out to say how wide it is:
$ hex2hex 'ff & 0f'
f
$ hex2hex --word 2 'ff & 0f'
000f
This is the same --word that groups bytes for a byte-order swap, because it is
the same claim about the same value — "this is a four-byte word". The two
compose:
$ ledec2hex --word 4 '1+1'
2000000
A negative reaching something with no way to say so is an error, not a magnitude quietly standing in:
$ hex2char '1-5'
hex2char: '1-5' is negative, and char cannot express that; add --word N to wrap it to N bytes
What is and is not an expression
An operator character is the whole trigger, and every one of them was a hard error before this existed — so nothing that worked has changed meaning. A newline or a comma still separates values; a space does not:
$ hex2dec ff 10 # no operator: two numbers, as always
255
16
$ hex2hex '1+1, 2+2' # two expressions
2
4
Only a positional source has expressions. char2hex '1+1' is still the hex
of those three characters, because char has no digits to compute with. And
--strict turns expression math off entirely, alongside the other leniencies,
for when the input is meant to be pure data.
Expressions are per line from a pipe, so printf '1+1\n2+2\n' | hex2hex gives
two answers.
Byte order
Either half of a name can carry an le or be prefix. This is the part that
matters for security work: a value pulled out of a memory dump, a ROP gadget
address, a struct field read off the wire.
$ lehex2dec 0d0c0b0a # 0x0a0b0c0d
168496141
$ lehex2hex 0d0c0b0a # the pure byte swap
0a0b0c0d
$ hex2lebin ff00
0000000011111111
$ lehex2char 6f6c6c6568
hello
be is the default written down — it changes nothing, and exists so a script
can be explicit rather than rely on a default it cannot see. Only the le
names are installed as commands, but every combination resolves through f2t,
so f2t leb642hex works whether or not you ran from2to install.
A byte order only means something when the byte width is known, and that width comes from how the source wrote the value:
| Source | Width | |
|---|---|---|
char, b64, b32 |
the run of bytes is its own width | reads to end of input |
hex, bin |
byte-aligned, so the digits you wrote are a byte count | lehex2dec ff is 255 — one byte has no byte order |
dec, oct, everything else |
none of their own | an error until --word says how wide |
That last row is deliberate. 255 is one byte and 256 is two, so inferring a width from the value would make the word size depend on the value — which is how a swap silently hands you the wrong address:
$ ledec2hex 258
ledec2hex: dec carries no byte width of its own; say how wide the word is: --word 4
Numbers swap per token: lehex2hex 0102 0304 is two results, not a
re-reading of four bytes. An odd hex digit count is an error rather than a
guess, since there is no byte boundary to swap on — -k downgrades that to a
warning like any other bad token.
--word
Groups the bytes into fixed-size words and reverses within each, leaving the order of the words alone. This is a hexdump off a 32-bit little-endian machine:
$ lehex2hex --word 4 0d0c0b0a78563412
0a0b0c0d12345678
$ dec2lehex --word 4 1234
d2040000
--word counts bytes of the value; --width counts digits of the output.
They are independent and may both be given.
Two things worth knowing in advance. -W strips the leading zeros of the
rendered result, which after a swap are the zeros that used to be trailing, so
lehex2hex -W 0100 is 1 rather than 0001. And reversing UTF-8 bytes is not
reversing characters — char2lehex é gives a9c3, which is no longer valid
UTF-8.
Escaped hex: esc
The \x48\x65 notation, in both directions. This is the format payloads are
written in, and the one you paste back out of them:
$ char2esc hello
\x68\x65\x6c\x6c\x6f
$ esc2char '\x66\x6c\x61\x67'
flag
Every byte is escaped on the way out, never "printable as-is, the rest escaped". Uniform output round-trips and never leaves you wondering whether a backslash was data.
Reading is deliberately more forgiving than writing, because what you copy out of an exploit script is mixed — literal text with escapes threaded through it, and every byte of it meant:
$ esc2hex 'GET /\r\n'
474554202f0d0a
The short escapes \n \r \t \0 \\ are understood on the way in. --strict
accepts nothing but \xNN.
It composes with byte order, which is where it earns its place — this is
p32(1234) from pwntools, and it agrees with struct.pack('<I', 1234) byte
for byte:
$ dec2leesc --word 4 1234
\xd2\x04\x00\x00
whatbase recognizes it on sight, since nothing else looks like it:
$ whatbase '\x68\x65\x6c\x6c\x6f'
esc hello
A backslash alone is not enough, though. \t and \n are a tab and a newline
in most languages, which would make every Windows path (C:\temp\new) read as
escaped bytes — so a real \xNN is required before esc is even a candidate.
char: bytes or code points?
hex2char reads bytes, because hex digits come in byte-sized pairs.
dec2char reads Unicode code points, because decimal digits do not:
$ hex2char 68656c6c6f
hello
$ dec2char 128512
😀
-b forces bytes, -c forces code points.
Flags
-w, --width N pad each result to N digits
-W, --no-width drop the natural digit width
-g, --group N group output digits in runs of N
-u, --upper uppercase the output
-x, --prefix prefix with 0x / 0b / 0o
-d, --delim S separator between results (\n, \t escapes work)
-j, --join no separator
-b, --bytes treat char as raw bytes
-c, --codepoint treat char as Unicode code points
--word BYTES bytes per word, for the le/be names
-k, --keep-going skip tokens that fail instead of stopping
-q, --quiet suppress warnings
-n, --no-newline no trailing newline
--strict reject prefixes, _ spacers and unpadded base64
A bad digit is an error and exits 1. -k downgrades it to a warning on stderr,
skips the token, and exits 0 — you asked it to keep going.
--strict also turns off expression math, and --word doubles
as the register width there.
Arbitrary size
Numbers are not limited to 64 or 128 bits:
$ dec2hex 340282366920938463463374607431768211456
100000000000000000000000000000000
Conversions between power-of-two bases never build a bignum at all — they are bit regrouping — so a multi-megabyte hex blob converts in one linear pass.
Building
$ cargo test
$ cargo build --release
Minimum supported Rust version is 1.85, the floor for edition 2024. CI holds that honest with a job pinned to it, so a newer language feature slipping in is a build failure rather than a surprise for someone on an older toolchain.
A man page lives at man/from2to.1; a test fails if it drifts
behind the flags. CHANGELOG.md records what changed between releases.
License
MIT — see LICENSE.