xonsh is a shell language and command prompt. Unlike other shells, xonsh is based on Python, with additional syntax added that makes calling subprocess commands, manipulating the environment, and dealing with the file system easy. The xonsh command prompt gives users interactive access to the xonsh language.

While all Python code is also xonsh, not all Bash code can be used in xonsh. That would defeat the purpose, and Python is better anyway! Still, xonsh is Bash-wards compatible in the ways that matter, such as for running commands, reading in the Bash environment, and utilizing Bash tab completion.

The purpose of this tutorial is to teach you xonsh. There are many excellent guides out there for learning Python, and this will not join their ranks. Similarly, you’d probably get the most out of this tutorial if you have already used a command prompt or interactive interpreter.

Let’s dive in!

Starting xonsh

Assuming you have successfully installed xonsh (see http://xon.sh), you can start up the xonsh interpreter via the xonsh command. Suppose you are in a lesser terminal:

$ xonsh
snail@home ~ $

Now we are in a xonsh shell. Our username happens to be snail, our hostname happens to be home, and we are in our home directory (~). Alternatively, you can setup your terminal emulator (xterm, gnome-terminal, etc) to run xonsh automatically when it starts up. This is recommended.


The xonsh language is based on Python, and the xonsh shell uses Python to interpret any input it receives. This makes simple things, like arithmetic, simple:

>>> 1 + 1


From here on we’ll be using >>> to prefix (or prompt) any xonsh input. This follows the Python convention and helps trick syntax highlighting, though $ is more traditional for shells.

Since this is just Python, we are able import modules, print values, and use other built-in Python functionality:

>>> import sys
>>> print(sys.version)
3.4.2 |Continuum Analytics, Inc.| (default, Oct 21 2014, 17:16:37)
[GCC 4.4.7 20120313 (Red Hat 4.4.7-1)]

We can also create and use literal data types, such as ints, floats, lists, sets, and dictionaries. Everything that you are used to if you already know Python is there:

>>> d = {'xonsh': True}
>>> d.get('bash', False)

The xonsh shell also supports multi-line input for more advanced flow control. The multi-line mode is automatically entered whenever the first line of input is not syntactically valid on its own. Multi-line mode is then exited when enter (or return) is pressed when the cursor is in the first column.

>>> if True:
...     print(1)
... else:
...     print(2)

Flow control, of course, includes loops.

>>> for i, x in enumerate('xonsh'):
...     print(i, x)
0 x
1 o
2 n
3 s
4 h

We can also define and call functions and classes. I’ll mostly spare you the details, but this is pretty cool:

>>> def f():
...     return "xonsh"
>>> f()

For easier indentation, Shift+Tab will enter 4 spaces. And that about wraps it up for the basics section. It is just like Python.

Environment Variables

Environment variables are written as $ followed by a name. For example, $HOME, $PWD, and $PATH.

>>> $HOME

You can set (and export) environment variables like you would set any other variable in Python. The same is true for deleting them too.

>>> $GOAL = 'Become the Lord of the Files'
>>> print($GOAL)
Become the Lord of the Files
>>> del $GOAL

Very nice.

The Environment Itself ${...}

All environment variables live in the built-in ${...} (aka __xonsh_env__) mapping. You can access this mapping directly, but in most situations, you shouldn’t need to.

If you want for example to check if an environment variable is present in your current session (say, in your awesome new xonsh script) you can use the membership operator:

>>> 'HOME' in ${...}

To get information about a specific enviroment variable you can use the help() method.

>>> ${...}.help('XONSH_DEBUG')

One helpful method on the ${...} is swap(). It can be used to temporarily set an environment variable:

>>> with ${...}.swap(SOMEVAR='foo'):
...     echo $SOMEVAR
>>> echo $SOMEVAR


Environment Lookup with ${<expr>}

The $NAME is great as long as you know the name of the environment variable you want to look up. But what if you want to construct the name programmatically, or read it from another variable? Enter the ${} operator.


In Bash, $NAME and ${NAME} are syntactically equivalent. In xonsh, they have separate meanings.

We can place any valid Python expression inside of the curly braces in ${<expr>}. This result of this expression will then be used to look up a value in the environment. Here are a couple of examples in action:

>>> x = 'USER'
>>> ${x}
>>> ${'HO' + 'ME'}

Not bad, xonsh, not bad.

Environment Types

Like other variables in Python, environment variables have a type. Sometimes this type is imposed based on the variable name. The current rules are pretty simple:

  • \w*PATH: any variable whose name ends in PATH is a list of strings.
  • \w*DIRS: any variable whose name ends in DIRS is a list of strings.
  • XONSH_HISTORY_SIZE: this variable is an int.
  • CASE_SENSITIVE_COMPLETIONS: this variable is a boolean.

xonsh will automatically convert back and forth to untyped (string-only) representations of the environment as needed (mostly by subprocess commands). When in xonsh, you’ll always have the typed version. Here are a couple of PATH examples:

>>> $PATH
['/home/snail/.local/bin', '/home/snail/sandbox/bin',
'/home/snail/miniconda3/bin', '/usr/local/bin', '/usr/local/sbin',
'/usr/bin', '/usr/sbin', '/bin', '/sbin', '.']
['/home/snail/.local/lib', '']

Also note that any Python object can go into the environment. It is sometimes useful to have more sophisticated types, like functions, in the environment. There are handful of environment variables that xonsh considers special. They can be seen on the Environment Variables page.


In subprocess mode, referencing an undefined environment variable will produce an empty string. In Python mode, however, a KeyError will be raised if the variable does not exist in the environment.

Running Commands

As a shell, xonsh is meant to make running commands easy and fun. Running subprocess commands should work like any other in any other shell.

>>> echo "Yoo hoo"
Yoo hoo
>>> cd xonsh
>>> ls
build  docs     README.rst  setup.py  xonsh           __pycache__
dist   license  scripts     tests     xonsh.egg-info
>>> dir scripts
xonsh  xonsh.bat
>>> git status
On branch master
Your branch is up-to-date with 'origin/master'.
Changes not staged for commit:
  (use "git add <file>..." to update what will be committed)
  (use "git checkout -- <file>..." to discard changes in working directory)

    modified:   docs/tutorial.rst

no changes added to commit (use "git add" and/or "git commit -a")
>>> exit

This should feel very natural.

Python-mode vs Subprocess-mode

It is sometimes helpful to make the distinction between lines that operate in pure Python mode and lines that use shell-specific syntax, edit the execution environment, and run commands. Unfortunately, it is not always clear from the syntax alone what mode is desired. This ambiguity stems from most command line utilities looking a lot like Python operators.

Take the case of ls -l. This is valid Python code, though it could have also been written as ls - l or ls-l. So how does xonsh know that ls -l is meant to be run in subprocess-mode?

For any given line that only contains an expression statement (expr-stmt, see the Python AST docs for more information), if all the names cannot be found as current variables xonsh will try to parse the line as a subprocess command instead. In the above, if ls and l are not variables, then subprocess mode will be attempted. If parsing in subprocess mode fails, then the line is left in Python-mode.

In the following example, we will list the contents of the directory with ls -l. Then we’ll make new variable names ls and l and then subtract them. Finally, we will delete ls and l and be able to list the directories again.

>>> # this will be in subproc-mode, because ls doesn't exist
>>> ls -l
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh
>>> # set ls and l variables to force python-mode
>>> ls = 44
>>> l = 2
>>> ls -l
>>> # deleting ls will return us to supbroc-mode
>>> del ls
>>> ls -l
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh

The determination between Python- and subprocess-modes is always done in the safest possible way. If anything goes wrong, it will favor Python-mode. The determination between the two modes is done well ahead of any execution. You do not need to worry about partially executed commands - that is impossible.

If you absolutely want to run a subprocess command, you can always force xonsh to do so with the syntax that we will see in the following sections.


Single or double quotes can be used to remove the special meaning of certain characters or words to xonsh. If a subprocess command contains characters that collide with xonsh syntax then quotes must be used to force xonsh to not interpret them.

>>> echo ${
SyntaxError: <xonsh-code>:1:5: ('code: {',)
echo ${
>>> echo '${'


There is no notion of an escaping character in xonsh like the backslash (\) in bash.

Captured Subprocess with $() and !()

The $(<expr>) operator in xonsh executes a subprocess command and captures some information about that command.

The $() syntax captures and returns the standard output stream of the command as a Python string. This is similar to how $() performs in Bash. For example,

>>> $(ls -l)
'total 0\n-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh\n'

The !() syntax captured more information about the command, as an instance of a class called CommandPipeline. This object contains more information about the result of the given command, including the return code, the process id, the standard output and standard error streams, and information about how input and output were redirected. For example:

>>> !(ls nonexistent_directory)
CommandPipeline(stdin=<_io.BytesIO object at 0x7f1948182bf8>, stdout=<_io.BytesIO object at 0x7f1948182af0>, stderr=<_io.BytesIO object at 0x7f19483a6200>, pid=26968, returncode=2, args=['ls', 'nonexistent_directory'], alias=['ls', '--color=auto', '-v'], stdin_redirect=['<stdin>', 'r'], stdout_redirect=[9, 'wb'], stderr_redirect=[11, 'w'], timestamps=[1485235484.5016758, None], executed_cmd=['ls', '--color=auto', '-v', 'nonexistent_directory'], input=None, output=, errors=None)

This object will be “truthy” if its return code was 0, and it is equal (via ==) to its return code. It also hashes to its return code. This allows for some interesting new kinds of interactions with subprocess commands, for example:

def check_file(file):
    if !(test -e @(file)):
        if !(test -f @(file)) or !(test -d @(file)):
            print("File is a regular file or directory")
            print("File is not a regular file or directory")
        print("File does not exist")

def wait_until_google_responds():
    while not !(ping -c 1 google.com):
        sleep 1

If you iterate over the CommandPipeline object, it will yield lines of its output. Using this, you can quickly and cleanly process output from commands. Additionally, these objects expose a method itercheck, which behaves the same as the built-in iterator but raises XonshCalledProcessError if the process had a nonzero return code.

def get_wireless_interface():
    """Returns devicename of first connected wifi, None otherwise"""
    for line in !(nmcli device):
        dev, typ, state, conn_name = line.split(None, 3)
        if typ == 'wifi' and state == 'connected':
            return dev

def grep_path(path, regexp):
    """Recursively greps `path` for perl `regexp`

    Returns a dict of 'matches' and 'failures'.
    Matches are files that contain the given regexp.
    Failures are files that couldn't be scanned.
    matches = []
    failures = []

        for match in !(grep -RPl @(regexp) @(str(path))).itercheck():
    except XonshCalledProcessError as error:
        for line in error.stderr.split('\n'):
            if not line.strip():
            filename = line.split('grep: ', 1)[1].rsplit(':', 1)[0]
    return {'matches': matches, 'failures': failures}

The $() and !() operators are expressions themselves. This means that we can assign the results to a variable or perform any other manipulations we want.

>>> x = $(ls -l)
>>> print(x.upper())
>>> y = !(ls -l)
>>> print(y.returncode)
>>> print(y.rtn)  # alias to returncode


Job control is not implemented for captured subprocesses.

While in subprocess-mode or inside of a captured subprocess, we can always still query the environment with $NAME variables or the ${} syntax, or inject Python values with the @() operator:

>>> $(echo $HOME)

Uncaptured Subprocess with $[] and ![]

Uncaptured subprocesses are denoted with the $[] and ![] operators. They are the same as $() captured subprocesses in almost every way. The only difference is that the subprocess’s stdout passes directly through xonsh and to the screen. The return value of $[] is always None.

In the following, we can see that the results of $[] are automatically printed, and the return value is not a string.

>>> x = $[ls -l]
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh
>>> x is None

Previously when we automatically entered subprocess-mode, uncaptured subprocesses were used. Thus ls -l and $[ls -l] are usually equivalent.

The ![] operator is similar to the !() in that it returns an object containing information about the result of executing the given command. However, its standard output and standard error streams are directed to the terminal, and the resulting object is not displayed. For example

>>> x = ![ls -l] and ![echo "hi"]
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh

Python Evaluation with @()

The @(<expr>) operator form works in subprocess mode, and will evaluate arbitrary Python code. The result is appended to the subprocess command list. If the result is a string, it is appended to the argument list. If the result is a list or other non-string sequence, the contents are converted to strings and appended to the argument list in order. If the result in the first position is a function, it is treated as an alias (see the section on Aliases below), even if it was not explicitly added to the aliases mapping. Otherwise, the result is automatically converted to a string. For example,

>>> x = 'xonsh'
>>> y = 'party'
>>> echo @(x + ' ' + y)
xonsh party
>>> echo @(2+2)
>>> echo @([42, 'yo'])
42 yo
>>> echo "hello" | @(lambda a, s=None: s.read().strip() + " world\n")
hello world

This syntax can be used inside of a captured or uncaptured subprocess, and can be used to generate any of the tokens in the subprocess command list.

>>> out = $(echo @(x + ' ' + y))
>>> out
'xonsh party\n'
>>> @("ech" + "o") "hey"

Thus, @() allows us to create complex commands in Python-mode and then feed them to a subprocess as needed. For example:

for i in range(20):
    $[touch @('file%02d' % i)]

Command Substitution with @$()

A common use of the @() and $() operators is allowing the output of a command to replace the command itself (command substitution): @([i.strip() for i in $(cmd).split()]). Xonsh offers a short-hand syntax for this operation: @$(cmd).

Consider the following example:

>>> # this returns a string representing stdout
>>> $(which ls)
'ls --color=auto\n'

>>> # this attempts to run the command, but as one argument
>>> # (looks for 'ls --color=auto\n' with spaces and newline)
>>> @($(which ls).strip())
xonsh: subprocess mode: command not found: ls --color=auto

>>> # this actually executes the intended command
>>> @([i.strip() for i in $(which ls).split()])
some_file  some_other_file

>>> # this does the same thing, but is much more concise
>>> @$(which ls)
some_file  some_other_file

Nesting Subprocesses

Though I am begging you not to abuse this, it is possible to nest the subprocess operators that we have seen so far ($(), $[], ${}, @(), @$()). An instance of ls -l that is on the wrong side of the border of the absurd is shown below:

>>> $[@$(which @($(echo ls).strip())) @('-' + $(printf 'l'))]
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh

With great power, and so forth...


Nesting these subprocess operators inside of $() and/or $[] works because the contents of those operators are executed in subprocess mode. Since @() and ${} run their contents in Python mode, it is not possible to nest other subprocess operators inside of them.


In subprocess-mode, xonsh allows you to use the | character to pipe together commands as you would in other shells.

>>> env | uniq | sort | grep PATH

This is only available in subprocess-mode because | is otherwise a Python operator. If you are unsure of what pipes are, there are many great references out there. You should be able to find information on StackOverflow or Google.

Logical Subprocess And

Subprocess-mode also allows you to use the and operator to chain together subprocess commands. The truth value of a command is evaluated as whether its return code is zero (i.e. proc.returncode == 0). Like in Python, if the command evaluates to False, subsequent commands will not be executed. For example, suppose we want to lists files that may or may not exist:

>>> touch exists
>>> ls exists and ls doesnt
/bin/ls: cannot access doesnt: No such file or directory

However, if you list the file that doesn’t exist first, you would have only seen the error:

>>> ls doesnt and ls exists
/bin/ls: cannot access doesnt: No such file or directory

Also, don’t worry. Xonsh directly translates the && operator into and for you. It is less Pythonic, of course, but it is your shell!

Logical Subprocess Or

Much like with and, you can use the or operator to chain together subprocess commands. The difference, to be certain, is that subsequent commands will be executed only if the if the return code is non-zero (i.e. a failure). Using the file example from above:

>>> ls exists or ls doesnt

This doesn’t even try to list a non-existent file! However, if you list the file that doesn’t exist first, you will see the error and then the file that does exist:

>>> ls doesnt or ls exists
/bin/ls: cannot access doesnt: No such file or directory

Never fear! Xonsh also directly translates the || operator into or, too. Your muscle memory is safe now, here with us.

Input/Output Redirection

xonsh also allows you to redirect stdin, stdout, and/or stderr. This allows you to control where the output of a command is sent, and where it receives its input from. xonsh has its own syntax for these operations, but, for compatibility purposes, xonsh also support Bash-like syntax.

The basic operations are “write to” (>), “append to” (>>), and “read from” (<). The details of these are perhaps best explained through examples.


The target of the redirection should be separated by a space, otherwise xonsh will raise a SyntaxError.

Redirecting stdout

All of the following examples will execute COMMAND and write its regular output (stdout) to a file called output.txt, creating it if it does not exist:

>>> COMMAND > output.txt
>>> COMMAND out> output.txt
>>> COMMAND o> output.txt
>>> COMMAND 1> output.txt # included for Bash compatibility

These can be made to append to output.txt instead of overwriting its contents by replacing > with >> (note that >> will still create the file if it does not exist).

Redirecting stderr

All of the following examples will execute COMMAND and write its error output (stderr) to a file called errors.txt, creating it if it does not exist:

>>> COMMAND err> errors.txt
>>> COMMAND e> errors.txt
>>> COMMAND 2> errors.txt # included for Bash compatibility

As above, replacing > with >> will cause the error output to be appended to errors.txt, rather than replacing its contents.

Combining Streams

It is possible to send all of COMMAND‘s output (both regular output and error output) to the same location. All of the following examples accomplish that task:

>>> COMMAND all> combined.txt
>>> COMMAND a> combined.txt
>>> COMMAND &> combined.txt # included for Bash compatibility

It is also possible to explicitly merge stderr into stdout so that error messages are reported to the same location as regular output. You can do this with the following syntax:

>>> COMMAND err>out
>>> COMMAND err>o
>>> COMMAND e>out
>>> COMMAND e>o
>>> COMMAND 2>&1  # included for Bash compatibility

This merge can be combined with other redirections, including pipes (see the section on Pipes above):

>>> COMMAND err>out | COMMAND2
>>> COMMAND e>o > combined.txt

It is worth noting that this last example is equivalent to: COMMAND a> combined.txt

Similarly, you can also send stdout to stderr with the following syntax:

>>> COMMAND out>err
>>> COMMAND out>e
>>> COMMAND o>err
>>> COMMAND o>e
>>> COMMAND 1>&2  # included for Bash compatibility

Redirecting stdin

It is also possible to have a command read its input from a file, rather than from stdin. The following examples demonstrate two ways to accomplish this:

>>> COMMAND < input.txt
>>> < input.txt COMMAND

Combining I/O Redirects

It is worth noting that all of these redirections can be combined. Below is one example of a complicated redirect.

>>> COMMAND1 e>o < input.txt | COMMAND2 > output.txt e>> errors.txt

This line will run COMMAND1 with the contents of input.txt fed in on stdin, and will pipe all output (stdout and stderr) to COMMAND2; the regular output of this command will be redirected to output.txt, and the error output will be appended to errors.txt.

Background Jobs

Typically, when you start a program running in xonsh, xonsh itself will pause and wait for that program to terminate. Sometimes, though, you may want to continue giving commands to xonsh while that program is running. In subprocess mode, you can start a process “in the background” (i.e., in a way that allows continued use of the shell) by adding an ampersand (&) to the end of your command. Background jobs are very useful when running programs with graphical user interfaces.

The following shows an example with emacs.

>>> emacs &

Note that the prompt is returned to you after emacs is started.

Job Control

If you start a program in the foreground (with no ampersand), you can suspend that program’s execution and return to the xonsh prompt by pressing Control-Z. This will give control of the terminal back to xonsh, and will keep the program paused in the background.


Suspending processes via Control-Z is not yet supported when running on Windows.

To unpause the program and bring it back to the foreground, you can use the fg command. To unpause the program have it continue in the background (giving you continued access to the xonsh prompt), you can use the bg command.

You can get a listing of all currently running jobs with the jobs command.

Each job has a unique identifier (starting with 1 and counting upward). By default, the fg and bg commands operate on the job that was started most recently. You can bring older jobs to the foreground or background by specifying the appropriate ID; for example, fg 1 brings the job with ID 1 to the foreground. Additionally, specify “+” for the most recent job and “-” for the second most recent job.

String Literals in Subprocess-mode

Strings can be used to escape special characters in subprocess-mode. The contents of the string are passed directly to the subprocess command as a single argument. So whenever you are in doubt, or if there is a xonsh syntax error because of a filename, just wrap the offending portion in a string.

A common use case for this is files with spaces in their names. This detestable practice refuses to die. “No problem!” says xonsh, “I have strings.” Let’s see it go!

>>> touch "sp ace"
>>> ls -l
total 0
-rw-rw-r-- 1 snail snail 0 Mar  8 17:50 sp ace
-rw-rw-r-- 1 snail snail 0 Mar  8 15:46 xonsh

By default, the name of an environment variable inside a string will be replaced by the contents of that variable (in subprocess mode only). For example:

>>> print("my home is $HOME")
my home is $HOME
>>> echo "my home is $HOME"
my home is /home/snail

You can avoid this expansion within a particular command by forcing the strings to be evaluated in Python mode using the @() syntax:

>>> echo "my home is $HOME"
my home is /home/snail
>>> echo @("my home is $HOME")
my home is $HOME

You can also disable environment variable expansion completely by setting $EXPAND_ENV_VARS to False.

Filename Globbing with *

Filename globbing with the * character is also allowed in subprocess-mode. This simply uses Python’s glob module under-the-covers. See there for more details. As an example, start with a lovely bunch of xonshs:

>>> touch xonsh conch konk quanxh
>>> ls
conch  konk  quanxh  xonsh
>>> ls *h
conch  quanxh  xonsh
>>> ls *o*
conch  konk  xonsh

This is not available in Python-mode because multiplication is pretty important.

Advanced Path Search with Backticks

xonsh offers additional ways to find path names beyond regular globbing, both in Python mode and in subprocess mode.

Regular Expression Globbing

If you have ever felt that normal globbing could use some more octane, then regex globbing is the tool for you! Any string that uses backticks (`) instead of quotes (', ") is interpreted as a regular expression to match filenames against. Like with regular globbing, a list of successful matches is returned. In Python-mode, this is just a list of strings. In subprocess-mode, each filename becomes its own argument to the subprocess command.

Let’s see a demonstration with some simple filenames:

>>> touch a aa aaa aba abba aab aabb abcba
>>> ls `a(a+|b+)a`
aaa  aba  abba
>>> print(`a(a+|b+)a`)
['aaa', 'aba', 'abba']
>>> len(`a(a+|b+)a`)

This same kind of search is performed if the backticks are prefaced with r. So the following expresions are equivalent: `test` and r`test`.

Other than the regex matching, this functions in the same way as normal globbing. For more information, please see the documentation for the re module in the Python standard library.


In Xonsh, the meaning of backticks is very different from their meaning in Bash. In Bash, backticks mean to run a captured subprocess ($() in Xonsh).

Normal Globbing

In subprocess mode, normal globbing happens without any special syntax. However, the backtick syntax has an additional feature: it is available inside of Python mode as well as subprocess mode.

Similarly to regex globbing, normal globbing can be performed (either in Python mode or subprocess mode) by using the g``:

>>> touch a aa aaa aba abba aab aabb abcba
>>> ls a*b*
aab  aabb  aba  abba  abcba
>>> ls g`a*b*`
aab  aabb  aba  abba  abcba
>>> print(g`a*b*`)
['aab', 'aabb', 'abba', 'abcba', 'aba']
>>> len(g`a*b*`)

Custom Path Searches

In addition, if normal globbing and regular expression globbing are not enough, xonsh allows you to specify your own search functions.

A search function is defined as a function of a single argument (a string) that returns a list of possible matches to that string. Search functions can then be used with backticks with the following syntax: @<name>`test`

The following example shows the form of these functions:

>>> def foo(s):
...     return [i for i in os.listdir('.') if i.startswith(s)]
>>> @foo`aa`
['aa', 'aaa', 'aab', 'aabb']

Path Output

Using the p modifier with either regex or glob backticks changes the return type from a list of strings to a list of pathlib.Path objects:

>>> p`.*`
[Path('foo'), Path('bar')]
>>> [x for x in pg`**` if x.is_symlink()]

Path Literals

Path objects can be instantiated directly using p-string syntax. Path objects can be converted back to plain strings with str(), and this conversion is handled implicitly in subprocess mode.

>>> mypath = p'/foo/bar'
>>> mypath
>>> mypath.stem
>>> echo @(mypath)

Help & Superhelp with ? & ??

From IPython, xonsh allows you to inspect objects with question marks. A single question mark (?) is used to display the normal level of help. Double question marks (??) are used to display a higher level of help, called superhelp. Superhelp usually includes source code if the object was written in pure Python.

Let’s start by looking at the help for the int type:

>>> int?
Type:            type
String form:     <class 'int'>
Init definition: (self, *args, **kwargs)
int(x=0) -> integer
int(x, base=10) -> integer

Convert a number or string to an integer, or return 0 if no arguments
are given.  If x is a number, return x.__int__().  For floating point
numbers, this truncates towards zero.

If x is not a number or if base is given, then x must be a string,
bytes, or bytearray instance representing an integer literal in the
given base.  The literal can be preceded by '+' or '-' and be surrounded
by whitespace.  The base defaults to 10.  Valid bases are 0 and 2-36.
Base 0 means to interpret the base from the string as an integer literal.
>>> int('0b100', base=0)
<class 'int'>

Now, let’s look at the superhelp for the xonsh built-in that enables regex globbing:

>>> __xonsh_regexsearch__??
Type:         function
String form:  <function regexsearch at 0x7efc8b367d90>
File:         /usr/local/lib/python3.5/dist-packages/xonsh/built_ins.py
Definition:   (s)
def regexsearch(s):
    s = expand_path(s)
    return reglob(s)

<function xonsh.built_ins.regexsearch>

Note that both help and superhelp return the object that they are inspecting. This allows you to chain together help inside of other operations and ask for help several times in an object hierarchy. For instance, let’s get help for both the dict type and its key() method simultaneously:

>>> dict?.keys??
Type:            type
String form:     <class 'dict'>
Init definition: (self, *args, **kwargs)
dict() -> new empty dictionary
dict(mapping) -> new dictionary initialized from a mapping object's
    (key, value) pairs
dict(iterable) -> new dictionary initialized as if via:
    d = {}
    for k, v in iterable:
        d[k] = v
dict(**kwargs) -> new dictionary initialized with the name=value pairs
    in the keyword argument list.  For example:  dict(one=1, two=2)
Type:        method_descriptor
String form: <method 'keys' of 'dict' objects>
Docstring:   D.keys() -> a set-like object providing a view on D's keys
<method 'keys' of 'dict' objects>

Of course, for subprocess commands, you still want to use the man command.

Compile, Evaluate, & Execute

Like Python and Bash, xonsh provides built-in hooks to compile, evaluate, and execute strings of xonsh code. To prevent this functionality from having serious name collisions with the Python built-in compile(), eval(), and exec() functions, the xonsh equivalents all append an ‘x’. So for xonsh code you want to use the compilex(), evalx(), and execx() functions. If you don’t know what these do, you probably don’t need them.


Another important xonsh built-in is the aliases mapping. This is like a dictionary that affects how subprocess commands are run. If you are familiar with the Bash alias built-in, this is similar. Alias command matching only occurs for the first element of a subprocess command.

The keys of aliases are strings that act as commands in subprocess-mode. The values are lists of strings, where the first element is the command, and the rest are the arguments. You can also set the value to a string, in which case it will be converted to a list automatically with shlex.split.

For example, the following creates several aliases for the git version control software. Both styles (list of strings and single string) are shown:

>>> aliases['g'] = 'git status -sb'
>>> aliases['gco'] = 'git checkout'
>>> aliases['gp'] = ['git', 'pull']

If you were to run gco feature-fabulous with the above aliases in effect, the command would reduce to ['git', 'checkout', 'feature-fabulous'] before being executed.

Callable Aliases

Lastly, if an alias value is a function (or other callable), then this function is called instead of going to a subprocess command. Such functions may have one of the following signatures:

def mycmd0():
    """This form takes no arguments but may return output or a return code.
    return "some output."

def mycmd1(args):
    """This form takes a single argument, args. This is a list of strings
    representing the arguments to this command. Feel free to parse them
    however you wish!
    # perform some action.
    return 0

def mycmd2(args, stdin=None):
    """This form takes two arguments. The args list like above, as a well
    as standard input. stdin will be a file like object that the command
    can read from, if the user piped input to this command. If no input
    was provided this will be None.
    # do whatever you want! Anything you print to stdout or stderr
    # will be captured for you automatically. This allows callable
    # aliases to support piping.
    print('I go to stdout and will be printed or piped')

    # Note: that you have access to the xonsh
    # built-ins if you 'import builtins'.  For example, if you need the
    # environment, you could do to following:
    import builtins
    env = builtins.__xonsh_env__

    # The return value of the function can either be None,

    # a single string representing stdout
    return  'I am out of here'

    # or you can build up strings for stdout and stderr and then
    # return a (stdout, stderr) tuple. Both of these may be
    # either a str or None. Any results returned like this will be
    # concatenated with the strings printed elsewhere in the function.
    stdout = 'I commanded'
    stderr = None
    return stdout, stderr

    # Lastly, a 3-tuple return value can be used to include an integer
    # return code indicating failure (> 0 return code). In the previous
    # examples the return code would be 0/success.
    return (None, "I failed", 2)

def mycmd3(args, stdin=None, stdout=None):
    """This form has three parameters.  The first two are the same as above.
    The last argument represents the standard output.  This is a file-like
    object that the command may write too.
    # you can either use stdout
    stdout.write("Hello, ")
    # or print()!

def mycmd4(args, stdin=None, stdout=None, stderr=None):
    """The next form of subprocess callables takes all of the
    arguments shown above as well as the standard error stream.
    As with stdout, this is a write-only file-like object.
    # This form allows "streaming" data to stdout and stderr
    import time
    for i in range(5):
        print(i, file=stdout)

    # In this form, the return value should be a single integer
    # representing the "return code" of the alias (zero if successful,
    # non-zero otherwise)
    return 0

def mycmd5(args, stdin=None, stdout=None, stderr=None, spec=None):
    """Lastly, the final form of subprocess callables takes all of the
    arguments shown above as well as a subpcrocess specification
    SubprocSpec object. This holds many attributes that dictate how
    the command is being run.  For instance this can be useful for
    knowing if the process is captured by $() or !().
    import xonsh.proc
    if spec.captured in xonsh.proc.STDOUT_CAPTURE_KINDS:
        end = ''
        end = '\n'
    # Now we'll get a newline if the user is at the terminal, and no
    # newline if we are captured
    print('Hi Mom!', end=end)
    return 0

Adding, Modifying, and Removing Aliases

We can dynamically alter the aliases present simply by modifying the built-in mapping. Here is an example using a function value:

>>> def _banana(args, stdin=None):
...     return ('My spoon is tooo big!', None)
>>> aliases['banana'] = _banana
>>> banana
'My spoon is tooo big!'

To redefine an alias, simply assign a new function, here using a python lambda with keyword arguments:

>>> aliases['banana'] = lambda: "Banana for scale.\n"
>>> banana
Banana for scale.

Removing an alias is as easy as deleting the key from the alias dictionary:

>>> del aliases['banana']


Alias functions should generally be defined with a leading underscore. Otherwise, they may shadow the alias itself, as Python variables take precedence over aliases when xonsh executes commands.

Anonymous Aliases

As mentioned above, it is also possible to treat functions outside this mapping as aliases, by wrapping them in @(). For example:

>>> @(_banana)
'My spoon is tooo big!'
>>> echo "hello" | @(lambda args, stdin=None: stdin.read().strip() + ' ' + args[0] + '\n') world
hello world

Unthreadable Aliases

Usually, callable alias commands will be run in a separate thread so that they may be run in the background. However, some aliases may need to be executed on the thread that they were called from. This is mostly useful for debuggers and profilers. To make an alias run in the foreground, decorate its function with the xonsh.tools.unthreadable decorator.

from xonsh.tools import unthreadable

def _mycmd(args, stdin=None):
    return 'In your face!'

aliases['mycmd'] = _mycmd

Uncapturable Aliases

Also, callable aliases by default will be executed such that their output is captured (like most commands in xonsh that don’t enter alternate mode). However, some aliases may want to run alternate-mode commands themselves. Thus the callable alias can’t be captured without dire consequences (tm). To prevent this, you can declare a callable alias uncapturable. This is mostly useful for aliases that then open up text editors, pagers, or the like. To make an alias uncapturable, decorate its function with the xonsh.tools.uncapturable decorator. This is probably best used in conjunction with the unthreadable decorator. For example:

from xonsh.tools import unthreadable, uncapturable

def _myvi():
    vi my.txt

aliases['myvi'] = _myvi

Aliasing is a powerful way that xonsh allows you to seamlessly interact to with Python and subprocess.


If FOREIGN_ALIASES_OVERRIDE enviroment variable is False (the default) then foreign shell aliases that try to override xonsh aliases will be ignored. Setting of this enviroment variable must happen in the static configuration file $XONSH_CONFIG_DIR/config.json in the ‘env’ section.

Up, Down, Tab

The up and down keys search history matching from the start of the line, much like they do in the IPython shell.

Tab completion is present as well. By default, in Python-mode you are able to complete based on the variable names in the current builtins, globals, and locals, as well as xonsh languages keywords & operator, files & directories, and environment variable names. In subprocess-mode, you additionally complete on the names of executable files on your $PATH, alias keys, and full Bash completion for the commands themselves.

xonsh also provides a means of modifying the behavior of the tab completer. More detail is available on the Tab Completion page.

Customizing the Prompt

Customizing the prompt by modifying $PROMPT is probably the most common reason for altering an environment variable.


Note that the $PROMPT variable will never be inherited from a parent process (regardless of whether that parent is a foreign shell or an instance of xonsh).

The $PROMPT variable can be a string, or it can be a function (of no arguments) that returns a string. The result can contain keyword arguments, which will be replaced automatically:

>>> $PROMPT = '{user}@{hostname}:{cwd} > '
snail@home:~ > # it works!
snail@home:~ > $PROMPT = lambda: '{user}@{hostname}:{cwd} >> '
snail@home:~ >> # so does that!

By default, the following variables are available for use:

  • user: The username of the current user
  • hostname: The name of the host computer
  • cwd: The current working directory, you may use $DYNAMIC_CWD_WIDTH to set a maximum width for this variable and $DYNAMIC_CWD_ELISION_CHAR to set the character used in shortened path.
  • short_cwd: A shortened form of the current working directory; e.g., /path/to/xonsh becomes /p/t/xonsh
  • cwd_dir: The dirname of the current working directory, e.g. /path/to in /path/to/xonsh.
  • cwd_base: The basename of the current working directory, e.g. xonsh in /path/to/xonsh.
  • env_name: The name of active virtual environment, if any.
  • curr_branch: The name of the current git branch, if any.
  • branch_color: {BOLD_GREEN} if the current git branch is clean, otherwise {BOLD_RED}. This is yellow if the branch color could not be determined.
  • branch_bg_color: Like, {branch_color}, but sets a background color instead.
  • prompt_end: # if the user has root/admin permissions $ otherwise
  • current_job: The name of the command currently running in the foreground, if any.
  • vte_new_tab_cwd: Issues VTE escape sequence for opening new tabs in the current working directory on some linux terminals. This is not usually needed.
  • gitstatus: Informative git status, like [master|MERGING|+1…2], you may use $XONSH_GITSTATUS_* to customize the styling.

You can also color your prompt easily by inserting keywords such as {GREEN} or {BOLD_BLUE}. Colors have the form shown below:

  • NO_COLOR: Resets any previously used color codes

  • COLORNAME: Inserts a color code for the following basic colors, which come in regular (dark) and intense (light) forms:

  • #HEX: A # before a len-3 or len-6 hex code will use that hex color, or the nearest approximation that that is supported by the shell and terminal. For example, #fff and #fafad2 are both valid.

  • BACKGROUND_ may be added to the begining of a color name or hex color to set a background color. For example, BACKGROUND_INTENSE_RED and BACKGROUND_#123456 can both be used.

  • bg#HEX or BG#HEX are shortcuts for setting a background hex color. Thus you can set bg#0012ab or the uppercase version.

  • BOLD_ is a prefix qualifier that may be used with any foreground color. For example, BOLD_RED and BOLD_#112233 are OK!

  • UNDERLINE_ is a prefix qualifier that also may be used with any foreground color. For example, UNDERLINE_GREEN.

  • Or any other combination of qualifiers, such as BOLD_UNDERLINE_INTENSE_BLACK, which is the most metal color you can use!

You can make use of additional variables beyond these by adding them to the PROMPT_FIELDS environment variable. The values in this dictionary should be strings (which will be inserted into the prompt verbatim), or functions of no arguments (which will be called each time the prompt is generated, and the results of those calls will be inserted into the prompt). For example:

snail@home ~ $ $PROMPT_FIELDS['test'] = "hey"
snail@home ~ $ $PROMPT = "{test} {cwd} $ "
hey ~ $
hey ~ $ import random
hey ~ $ $PROMPT_FIELDS['test'] = lambda: random.randint(1,9)
3 ~ $
5 ~ $
2 ~ $
8 ~ $

Environment variables and functions are also available with the $ prefix. For example:

snail@home ~ $ $PROMPT = "{$LANG} >"
en_US.utf8 >

Note that some entries of the $PROMPT_FIELDS are not always applicable, for example, curr_branch returns None if the current directory is not in a repository. The None will be interpreted as an empty string.

But let’s consider a problem:

snail@home ~/xonsh $ $PROMPT = "{cwd_base} [{curr_branch}] $ "
xonsh [master] $ cd ..
~ [] $

We want the branch to be displayed in square brackets, but we also don’t want the brackets (and the extra space) to be displayed when there is no branch. The solution is to add a nested format string (separated with a colon) that will be invoked only if the value is not None:

snail@home ~/xonsh $ $PROMPT = "{cwd_base}{curr_branch: [{}]} $ "
xonsh [master] $ cd ..
~ $

The curly brackets act as a placeholder, because the additional part is an ordinary format string. What we’re doing here is equivalent to this expression:

" [{}]".format(curr_branch()) if curr_branch() is not None else ""

Executing Commands and Scripts

When started with the -c flag and a command, xonsh will execute that command and exit, instead of entering the command loop.

$ xonsh -c "echo @(7+3)"

Longer scripts can be run either by specifying a filename containing the script, or by feeding them to xonsh via stdin. For example, consider the following script, stored in test.xsh:

#!/usr/bin/env xonsh


print('removing files')
rm `file\d+.txt`


print('adding files')
# This is a comment
for i, x in enumerate("xonsh"):
    echo @(x) > @("file{0}.txt".format(i))

print($(ls).replace('\n', ' '))

This script could be run by piping its contents to xonsh:

$ cat test.xsh | xonsh
file0.txt  file1.txt  file2.txt  file3.txt  file4.txt  test_script.sh
removing files
adding files
file0.txt file1.txt file2.txt file3.txt file4.txt test_script.sh

or by invoking xonsh with its filename as an argument:

$ xonsh test.xsh
file0.txt  file1.txt  file2.txt  file3.txt  file4.txt  test_script.sh
removing files
adding files
file0.txt file1.txt file2.txt file3.txt file4.txt test_script.sh

xonsh scripts can also accept arguments. These arguments are made available to the script in two different ways:

  1. In either mode, as individual variables $ARG<n> (e.g., $ARG1)
  2. In Python mode only, as a list $ARGS

For example, consider a slight variation of the example script from above that operates on a given argument, rather than on the string 'xonsh' (notice how $ARGS and $ARG1 are used):

#!/usr/bin/env xonsh



print('removing files')
rm `file\d+.txt`


print('adding files')
# This is a comment
for i, x in enumerate($ARG1):
    echo @(x) > @("file{0}.txt".format(i))

print($(ls).replace('\n', ' '))
$ xonsh test2.xsh snails
['test_script.sh', 'snails']
file0.txt  file1.txt  file2.txt  file3.txt  file4.txt  file5.txt  test_script.sh
removing files
adding files
file0.txt file1.txt file2.txt file3.txt file4.txt file5.txt test_script.sh

$ echo @(' '.join($(cat @('file%d.txt' % i)).strip() for i in range(6)))
s n a i l s

Additionally, if the script should exit if a command fails, set the environment variable $RAISE_SUBPROC_ERROR = True at the top of the file. Errors in Python mode will already raise exceptions and so this is roughly equivalent to Bash’s set -e.

Furthermore, you can also toggle the ability to print source code lines with the trace on and trace off commands. This is roughly equivelent to Bash’s set -x or Python’s python -m trace, but you know, better.

Importing Xonsh (*.xsh)

You can import xonsh source files with the *.xsh file extension using the normal Python syntax. Say you had a file called mine.xsh, you could, therefore, perform a Bash-like source into your current shell with the following:

from mine import *

That’s All, Folks

To leave xonsh, hit Ctrl-D, type EOF, type quit, or type exit. On Windows, you can also type Ctrl-Z.

>>> exit

Now it is your turn.