Starting with 9.0, IPython will be able to use LLM providers to suggest code in the terminal. This requires a recent version of prompt_toolkit in order to allow multiline suggestions. There are currently a number of limitations, and feedback on the API is welcome.

Unlike many of IPython features, this is not enabled by default and requires multiple configuration options to be set to properly work:

• Set a keybinding to trigger LLM suggestions. Due to terminal limitations across platforms and emulators, it is difficult to provide a default keybinding. Note that not all keybindings are availables, in particular all the Ctrl-Enter, Alt-backslash and Ctrl-Shift-Enter are not available without integration with your terminal emulator.
• Chose a LLM provider, usually from Jupyter-AI. This will be the interface between IPython itself, and the LLM – that may be local or in on a server.
• Configure said provider with models, API keys, etc – this will depend on the provider, and you will have to refer to Jupyter-AI documentation, and/or your LLM documenatation.

While setting up IPython to use a real LLM, you can refer to examples/auto_suggest_llm.py that both provide an example of how to set up IPython to use a Fake LLM provider, this can help ensure that the full setup is working before switching to a real LLM provider.

You may want to refer on how to setup a keybinding in IPython, but in short you want to bind the IPython:auto_suggest.llm_autosuggestion command to a keybinding, and have it active only when the default buffer isi focused, and when using the NavigableSuggestions suggestter (this is the default suggestter, the one that is history and LLM aware). Thus the navigable_suggestions & default_buffer_focused filter should be used.

Usually Ctrl-Q on macos is an available shortcut, note that is does use Ctrl, and not Command.

The following example will bind Ctrl-Q to the llm_autosuggestion command, with the suggested filter:

c.TerminalInteractiveShell.shortcuts = [
    {
        "new_keys": ["c-q"],
        "command": "IPython:auto_suggest.llm_autosuggestion",
        "new_filter": "navigable_suggestions & default_buffer_focused",
        "create": True,
    },
]

Set the TerminalInteractiveShell.llm_provider_class trait to the fully qualified name of the Provider you like, when testing from inside the IPython source tree, you can use "examples.auto_suggest_llm.ExampleCompletionProvider" This will always stream an extract of the Little Prince by Antoine de Saint-Exupéry, and will not require any API key or real LLM.

In your configuration file adapt the following line to your needs:

c.TerminalInteractiveShell.llm_provider_class = "examples.auto_suggest_llm.ExampleCompletionProvider"

It the provider needs to be passed parameters at initialization, you can do so by setting the llm_construction_kwargs traitlet.

c.TerminalInteractiveShell.llm_constructor_kwargs = {"model": "skynet"}

This will depdend on the provider you chose, and you will have to refer to the provider documentation.

Extra configuration may be needed by setting environment variables, this will again depend on the provider you chose, and you will have to refer to the provider documentation.

The option c.TerminalInteractiveShell.llm_prefix_from_history controls the context the Provider gets when trying to complete. See the help of this options (ipython --help-all):

Fully Qualifed name of a function that takes an IPython history manager and
return a prefix to pass the llm provider in addition to the current buffer
text.

You can use:

 - no_prefix
 - input_history

As default value. `input_history` (default),  will use all the input history
of current IPython session

.. versionchanged:: 5.0

From IPython 5, prompts are produced as a list of Pygments tokens, which are tuples of (token_type, text). You can customise prompts by writing a method which generates a list of tokens.

There are four kinds of prompt:

• The in prompt is shown before the first line of input (default like In [1]:).
• The continuation prompt is shown before further lines of input (default like ...:).
• The rewrite prompt is shown to highlight how special syntax has been interpreted (default like ----->).
• The out prompt is shown before the result from evaluating the input (default like Out[1]:).

Custom prompts are supplied together as a class. If you want to customise only some of the prompts, inherit from :class:`IPython.terminal.prompts.Prompts`, which defines the defaults. The required interface is like this:

Prompt style definition. shell is a reference to the :class:`~.TerminalInteractiveShell` instance.

.. method:: in_prompt_tokens()
            continuation_prompt_tokens(self, width=None)
            rewrite_prompt_tokens()
            out_prompt_tokens()

   Return the respective prompts as lists of ``(token_type, text)`` tuples.

   For continuation prompts, *width* is an integer representing the width of
   the prompt area in terminal columns.

Here is an example Prompt class that will show the current working directory in the input prompt:

from IPython.terminal.prompts import Prompts, Token
import os

class MyPrompt(Prompts):
     def in_prompt_tokens(self):
         return [(Token, os.getcwd()),
                 (Token.Prompt, ' >>>')]

To set the new prompt, assign it to the prompts attribute of the IPython shell:

In [2]: ip = get_ipython()
   ...: ip.prompts = MyPrompt(ip)

/home/bob >>> # it works

See IPython/example/utils/cwd_prompt.py for an example of how to write extensions to customise prompts.

Inside IPython or in a startup script, you can use a custom prompts class by setting get_ipython().prompts to an instance of the class. In configuration, TerminalInteractiveShell.prompts_class may be set to either the class object, or a string of its full importable name.

To include invisible terminal control sequences in a prompt, use Token.ZeroWidthEscape as the token type. Tokens with this type are ignored when calculating the width.

Colours in the prompt are determined by the token types and the highlighting style; see below for more details. The tokens used in the default prompts are Prompt, PromptNum, OutPrompt and OutPromptNum.

.. versionchanged:: 9.0

IPython 9.0 changed almost all of the color handling, which is now referred to as themes. A Theme can do a bit more than purely colors, as it can handle bold, italic and basically any style that pygments support. Themes also support a number of Symbols, which allows you to – for example – change the shape of the arrow that mark the current frame and line numbers in the debugger and the tracebacks.

Most of the various IPython options that were used pre 9.0 have been renamed, with a exceptions a few, and most classes that deal with themes can, now take a theme_name parameter.

To reflect this, the --colors flag now is also aliased to --theme.

The default themes included are the same, except lowercase, for ease of typing.

'nocolor', 'neutral', 'linux', 'lightbg', 'gruvbox-dark', with the addition of 'pride' to celebrate the inclusively of this project (I welcome update to the pride theme as I'm not a designer myself).

In addition, the --theme=pride theme, is the first to make use of unicode symbols for the traceback separation line, and the debugger and traceback arrow, as well as making some use of bold, and italic formatting, and not limit itself to the 16 base ANSI colors.

We encourage you to contribute themes, and to distribute them, while currently you need to modify source code to add a theme, it should be possible to load theme from Json, Yaml, or any other declarative file type.

Since IPython 9.0, most of IPython internal code emit a sequence of (Token Type, string), which is fed through pygments, and a theme is mapping from those token types to a style. For example: Token.Prompt : '#ansired underline', or Token.Filename : 'bg:#A30262.

For simplicity, a theme can be derived from from a pygments style (which will give the basic code highlighting).

A theme can also define a few symbols (see the source for how), for example arrow_body, and arrow_head, can help customising line indicators.

On some systems, the default pager has problems with ANSI colour codes. To configure your default pager to allow these:

1. Set the environment PAGER variable to less.
2. Set the environment LESS variable to -r (plus any other options you always want to pass to less by default). This tells less to properly interpret control sequences, which is how color information is given to your terminal.

IPython can integrate with text editors in a number of different ways:

• Editors (such as (X)Emacs, vim and TextMate) can send code to IPython for execution.
• IPython's %edit magic command can open an editor of choice to edit a code block.

The %edit command (and its alias %ed) will invoke the editor set in your environment as :envvar:`EDITOR`. If this variable is not set, it will default to vi under Linux/Unix and to notepad under Windows. You may want to set this variable properly and to a lightweight editor which doesn't take too long to start (that is, something other than a new instance of Emacs). This way you can edit multi-line code quickly and with the power of a real editor right inside IPython.

You can also control the editor by setting :attr:`TerminalInteractiveShell.editor` in :file:`ipython_config.py`.

Paul Ivanov's vim-ipython provides powerful IPython integration for vim.

If you are a dedicated Emacs user, and want to use Emacs when IPython's %edit magic command is called you should set up the Emacs server so that new requests are handled by the original process. This means that almost no time is spent in handling the request (assuming an Emacs process is already running). For this to work, you need to set your EDITOR environment variable to 'emacsclient'. The code below, supplied by Francois Pinard, can then be used in your :file:`.emacs` file to enable the server:

(defvar server-buffer-clients)
(when (and (fboundp 'server-start) (string-equal (getenv "TERM") 'xterm))
  (server-start)
  (defun fp-kill-server-with-buffer-routine ()
    (and server-buffer-clients (server-done)))
  (add-hook 'kill-buffer-hook 'fp-kill-server-with-buffer-routine))

Thanks to the work of Alexander Schmolck and Prabhu Ramachandran, currently (X)Emacs and IPython get along very well in other ways.

With (X)EMacs >= 24, You can enable IPython in python-mode with:

(require 'python)
(setq python-shell-interpreter "ipython")

.. versionadded:: 8.11

You can modify, disable or modify keyboard shortcuts for IPython Terminal using :std:configtrait:`TerminalInteractiveShell.shortcuts` traitlet.

The list of shortcuts is available in the Configuring IPython :ref:`terminal-shortcuts-list` section.

.. versionchanged:: 5.0

Creating custom commands requires adding custom code to a :ref:`startup file <startup_files>`:

from IPython import get_ipython
from prompt_toolkit.enums import DEFAULT_BUFFER
from prompt_toolkit.keys import Keys
from prompt_toolkit.filters import HasFocus, HasSelection, ViInsertMode, EmacsInsertMode

ip = get_ipython()
insert_mode = ViInsertMode() | EmacsInsertMode()

def insert_unexpected(event):
    buf = event.current_buffer
    buf.insert_text('The Spanish Inquisition')
# Register the shortcut if IPython is using prompt_toolkit
if getattr(ip, 'pt_app', None):
    registry = ip.pt_app.key_bindings
    registry.add_binding(Keys.ControlN,
                     filter=(HasFocus(DEFAULT_BUFFER)
                             & ~HasSelection()
                             & insert_mode))(insert_unexpected)

Here is a second example that bind the key sequence j, k to switch to VI input mode to Normal when in insert mode:

from IPython import get_ipython
from prompt_toolkit.enums import DEFAULT_BUFFER
from prompt_toolkit.filters import HasFocus, ViInsertMode
from prompt_toolkit.key_binding.vi_state import InputMode

ip = get_ipython()

def switch_to_navigation_mode(event):
   vi_state = event.cli.vi_state
   vi_state.input_mode = InputMode.NAVIGATION

if getattr(ip, 'pt_app', None):
   registry = ip.pt_app.key_bindings
   registry.add_binding(u'j',u'k',
                        filter=(HasFocus(DEFAULT_BUFFER)
                                 & ViInsertMode()))(switch_to_navigation_mode)

For more information on filters and what you can do with the event object, see the prompt_toolkit docs.

In the Terminal IPython shell – which by default uses the prompt_toolkit interface, the semantic meaning of pressing the Enter key can be ambiguous. In some case Enter should execute code, and in others it should add a new line. IPython uses heuristics to decide whether to execute or insert a new line at cursor position. For example, if we detect that the current code is not valid Python, then the user is likely editing code and the right behavior is to likely to insert a new line. If the current code is a simple statement like ord('*'), then the right behavior is likely to execute. Though the exact desired semantics often varies from users to users.

As the exact behavior of Enter is ambiguous, it has been special cased to allow users to completely configure the behavior they like. Hence you can have enter always execute code. If you prefer fancier behavior, you need to get your hands dirty and read the prompt_toolkit and IPython documentation though. See :ghpull:`10500`, set the c.TerminalInteractiveShell.handle_return option and get inspiration from the following example that only auto-executes the input if it begins with a bang or a modulo character (! or %). To use the following code, add it to your IPython configuration:

def custom_return(shell):

    """This function is required by the API. It takes a reference to
    the shell, which is the same thing `get_ipython()` evaluates to.
    This function must return a function that handles each keypress
    event. That function, named `handle` here, references `shell`
    by closure."""

    def handle(event):

        """This function is called each time `Enter` is pressed,
        and takes a reference to a Prompt Toolkit event object.
        If the current input starts with a bang or modulo, then
        the input is executed, otherwise a newline is entered,
        followed by any spaces needed to auto-indent."""

        # set up a few handy references to nested items...

        buffer = event.current_buffer
        document = buffer.document
        text = document.text

        if text.startswith('!') or text.startswith('%'): # execute the input...

            buffer.accept_action.validate_and_handle(event.cli, buffer)

        else: # insert a newline with auto-indentation...

            if document.line_count > 1: text = text[:document.cursor_position]
            indent = shell.check_complete(text)[1]
            buffer.insert_text('\n' + indent)

            # if you just wanted a plain newline without any indentation, you
            # could use `buffer.insert_text('\n')` instead of the lines above

    return handle

c.TerminalInteractiveShell.handle_return = custom_return