NVDA 2024.2 Developer Guide


This guide provides information concerning NVDA development, including translation and the development of components for NVDA.

Add-on API stability

The NVDA Add-on API includes all NVDA internals, except symbols that are prefixed with an underscore.

The NVDA Add-on API changes over time, such as removals, deprecations and new features. Important changes to the API are announced on the NVDA API mailing list. Changes relevant to developers are also announced via the NVDA changes file. Any changes to the API policy outlined in this section will be conveyed via these two channels.

API breaking releases happen at most once per year, these are .1 releases, e.g. 2022.1. The API remains backwards compatible between breaking releases. API breaking changes should be considered relatively stable in the first beta: e.g. 2022.1.beta1.

API features may become deprecated over time. Deprecated API features may have a scheduled removal date, a future breaking release (e.g. 2022.1). Deprecations may also have no scheduled removal date, and will remain supported until it is no longer reasonable. Note, the roadmap for removals is 'best effort' and may be subject to change. Please open a GitHub issue if the described add-on API changes result in the API no longer meeting the needs of an add-on you develop or maintain.

A Note About Python

NVDA and its components are primarily written in the Python programming language. It is not the goal of this guide to teach you Python, though examples are provided through out this guide which will help to familiarise you with the Python syntax. Documentation and other resources related to the Python language can be found at www.python.org/


Some of NVDA is written in C++, e.g. nvdaHelper. For an overview of nvdaHelper, including how to configure Visual Studio to enable intellisense see the nvdaHelper readme


In order to support multiple languages/locales, NVDA must be translated and data specific to the locale must be provided. This section only includes information on custom NVDA file formats required for translation. Other items need to be translated, such as the NVDA user interface and documentation, but these use standard file formats. For complete documentation about translating NVDA, please see the Translating wiki page

Character Descriptions

Sometimes, it can be very difficult or even impossible to distinguish one character from another. For example, two characters might be pronounced the same way, even though they are actually different characters. To help users when this occurs, character descriptions can be provided which describe the character in a unique way.

Character descriptions can be provided for a locale in a file named characterDescriptions.dic in the directory for the locale. This is a UTF-8 encoded text file. Blank lines and lines beginning with a "#" character are ignored. All other lines should contain a character, followed by a tab, then one or more descriptions separated by tabs.

For example:

# This is a comment.
a   alpha
b   bravo

See the file locale\en\characterDescriptions.dic for a full example.

In most cases, the characters in this file should be a single lower case character. It is assumed that characters will have the same description regardless of their case, so upper case characters are converted to lower case before looking up their character descriptions.

Symbol Pronunciation

It is often useful to hear punctuation and other symbols pronounced as words when reading text, particularly when moving by character. Unfortunately, the pronunciation of symbols is inconsistent between speech synthesisers and many synthesisers do not speak many symbols and/or do not allow control over what symbols are spoken. Therefore, NVDA allows information about symbol pronunciation to be provided.

This is done for a locale by providing a file named symbols.dic in the directory for the locale. This is a UTF-8 encoded text file. Blank lines and lines beginning with a "#" character are ignored. All locales implicitly inherit the symbol information for English, though any of this information can be overridden.

The file contains two sections.

Defining Complex Symbols

The first section is optional and defines regular expression patterns for complex symbols. Complex symbols are symbols which aren't simply a character or sequence of characters, but instead require a more complicated match. An example is the full stop (.) sentence ending in English. The "." is used for multiple purposes, so a more complicated check is required to determine whether this refers to the end of a sentence.

The complex symbols section begins with the line:


Subsequent lines contain a textual identifier used to identify the symbol, a tab and the regular expression pattern for that symbol. For example:

. sentence ending   (?<=[^\s.])\.(?=[\"')\s]|$)
dates with .    \b(\d\d)\.(\d\d)\.(\d{2}|\d{4})\b

Again, the English symbols are inherited by all other locales, so you need not include any complex symbols already defined for English.

Defining Symbol Information

The second section provides information about when and how to pronounce all symbols. It begins with the line:


Subsequent lines should contain several fields separated by tabs. The only mandatory fields are the identifier and replacement. The default will be used for omitted fields. The fields are as follows:

Finally, a display name for the symbol can be provided in a comment after a tab at the end of the line. This will be shown to users when editing the symbol information and is especially useful for translators to define translated names for English complex symbols.

Here are some examples:

(   left paren  most

This means that the "(" character should be spoken as "left paren" only when the symbol level is set to most or higher; i.e. most or all.

,   comma   all always

This means that the "," character should be spoken as "comma" when the symbol level is set to all and that the character itself should always be preserved so that the synthesiser will pause appropriately.

. sentence ending   point   # . fin de phrase

This line appears in the French symbols.dic file. It means that the ". sentence ending" complex symbol should be spoken as "point". Level and preserve are not specified, so they will be taken from English. A display name is provided so that French users will know what the symbol represents.

dates with .    \1 point \2 point \3    all norep   # date avec points

This line appears in the French symbols.dic file. It means that the first, second, and third groups of the match will be included, separated by the word 'point'. The effect is thus to replace the dots from the date with the word 'point'.

Please see the file locale\en\symbols.dic for the English definitions which are inherited for all locales. This is also a good full example.



Plugins allow you to customize the way NVDA behaves overall or within a particular application. They are able to:

This section provides an introduction to developing plugins. Developers should consult the code documentation for a complete reference.

Types of Plugins

There are two types of plugins. These are:

If you wish to improve NVDA's access to a particular application, it is most likely you will want to write an App Module. In contrast, if you wish to add some overall functionality to NVDA (e.g. a script that announces current Wireless network strength while in any application), then a Global Plugin is what you want.

Both App Modules and Global Plugins share a common look and feel. They are both Python source files (with a .py extension), they both define a special class containing all events, scripts and bindings, and they both may define custom classes to access controls, text content and complex documents. However, they do differ in some ways.

Custom appModules and globalPlugins can be packaged into NVDA add-ons. This allows easy distribution, and provides a safe way for the user to install and uninstall the custom code. Please refer to the Add-ons section later on in this document.

In order to test the code while developing, you can place it in a special 'scratchpad' directory in your NVDA user configuration directory. You will also need to configure NVDA to enable loading of custom code from the Developer Scratchpad Directory, by enabling this in the Advanced category of NVDA's Settings dialog. The Advanced category also contains a button to easily open the Developer Scratchpad directory if enabled.

The following few sections will talk separately about App Modules and Global Plugins. After this point, discussion is again more general.

Basics of an App Module

App Module files have a .py extension, and in most cases should be named the same as either the main executable of the application for which you wish them to be used or the package inside a host executable. For example, an App Module for notepad would be called notepad.py, as notepad's main executable is called notepad.exe. To map a single App Module for multiple executables, or handle when an executable name violates the Python import rules, refer to Associating App Modules with an executable. For apps hosted inside host executables, see the section on app modules for hosted apps.

App Module files must be placed in the appModules subdirectory of an add-on, or of the scratchpad directory of the NVDA user configuration directory.

App Modules must define a class called AppModule, which inherits from appModuleHandler.AppModule. This class can then define event and script methods, gesture bindings and other code. This will all be covered in depth later.

NVDA loads an App Module for an application as soon as it notices the application is running. The App Module is unloaded once the application is closed or when NVDA is exiting.

Associating App Modules with an executable

As explained above, sometimes the default way of associating an App Module with an application is not flexible enough. Examples include:

In such cases you can distribute a small global plugin along with your App Module which maps it to the executable. For example to map the App Module named "time_app_mod" to the "time" executable the plugin may be written as follows:

import appModuleHandler
import globalPluginHandler

class GlobalPlugin(globalPluginHandler.GlobalPlugin):

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        appModuleHandler.registerExecutableWithAppModule("time", "time_app_mod")

    def terminate(self, *args, **kwargs):
        super().terminate(*args, **kwargs)

Example 1: An App Module that Beeps on Focus Change Events

The following example App Module makes NVDA beep each time the focus changes within the notepad application. This example shows you the basic layout of an App Module.

Copy and paste the code between (but not including) the start and end markers into a new text file called notepad.py, which should be saved in the AppModules subdirectory. Be very careful to keep all tabs and spaces intact.

Once saved in the correct location, either restart NVDA or choose Reload Plugins found under Tools in the NVDA menu.

Finally, open Notepad and move the focus around the application; e.g. move along the menu bar, open some dialog boxes, etc. You should hear beeps each time the focus changes. Note though that if you move outside of Notepad - for instance, to Windows Explorer - you do not hear beeps.

--- start ---
# Notepad App Module for NVDA
# Developer guide example 1

import appModuleHandler

class AppModule(appModuleHandler.AppModule):

    def event_gainFocus(self, obj, nextHandler):
        import tones
        tones.beep(550, 50)

--- end ---

This App Module file starts with two comment lines, which describe what the file is for.

It then imports the appModuleHandler module, so that the App Module then has access to the base AppModule class.

Next, it defines a class called AppModule, which is inherited from appModuleHandler.AppModule.

Inside this class, it defines 1 or more events, scripts or gesture bindings. In this example, it defines one event method for gainFocus events (event_gainFocus), which plays a short beep each time it is executed. The implementation of this event is not important for the purposes of this example. The most important part is the class itself. Events will be covered in greater detail later.

As with other examples in this guide, remember to delete the created app module when you are finished testing and then restart NVDA or reload plugins, so that original functionality is restored.

App modules for hosted apps

Some executables host various apps inside. These include javaw.exe for running various Java programs and wwahost.exe for some apps in Windows 8 and later.

If an app runs inside a host executable, the name of the app module must be the name as defined by the host executable, which can be found through AppModule.appName property. For example, an app module for a Java app named "test" hosted inside javaw.exe must be named test.py. For apps hosted inside wwahost, not only must the app module name be the name of the loaded app, but the app module must subclass the app module class found in wwahost.

Example 2: an app module for an app hosted by wwahost.exe

The following example is same as Notepad app module above except this is for an app hosted by wwahost.exe.

--- start ---
# wwahost/test App Module for NVDA
# Developer guide example 2

from nvdaBuiltin.appModules.wwahost import *

class AppModule(AppModule):

    def event_gainFocus(self, obj, nextHandler):
        import tones
        tones.beep(550, 50)

--- end ---

The biggest difference from Notepad app module is where wwahost app module comes from. As a built-in app module, wwahost can be imported from nvdaBuiltin.appModules.

Another difference is how the app module class is defined. As wwahost app module provides necessary infrastructure for apps hosted inside, you just need to subclass the wwahost AppModule class.

Basics of a Global Plugin

Global Plugin files have a .py extension, and should have a short unique name which identifies what they do.

Global plugin files must be placed in the globalPlugins subdirectory of an add-on, or of the scratchpad directory of the NVDA user configuration directory.

Global Plugins must define a class called GlobalPlugin, which inherits from globalPluginHandler.GlobalPlugin. This class can then define event and script methods, gesture bindings and other code. This will all be covered in depth later.

NVDA loads all global plugins as soon as it starts, and unloads them on exit.

Example 3: a Global Plugin Providing a Script to Announce the NVDA Version

The following example Global Plugin Allows you to press NVDA+shift+v while anywhere in the Operating System to find out NVDA's version. This example is only to show you the basic layout of a Global Plugin.

Copy and paste the code between (but not including) the start and end markers into a new text file with a name of example2.py, which should be saved in the globalPlugins subdirectory. Be very careful to keep all tabs and spaces intact.

Once saved in the right place, either restart NVDA or choose Reload Plugins found under Tools in the NVDA menu.

From anywhere, you can now press NVDA+shift+v to have NVDA's version spoken and brailled.

--- start ---
# Version announcement plugin for NVDA
# Developer guide example 3

import globalPluginHandler
from scriptHandler import script
import ui
import versionInfo

class GlobalPlugin(globalPluginHandler.GlobalPlugin):

    def script_announceNVDAVersion(self, gesture):

--- end ---

This Global Plugin file starts with two comment lines, which describe what the file is for.

It then imports the globalPluginHandler module, so that the Global Plugin has access to the base GlobalPlugin class.

It also imports a few other modules, namely ui, versionInfo and scriptHandler, which this specific plugin needs in order for it to perform the necessary actions to announce the version.

Next, it defines a class called GlobalPlugin, which is inherited from globalPluginHandler.GlobalPlugin.

Inside this class, it defines 1 or more events, scripts or gesture bindings. In this example, it defines a script method that performs the version announcement. The script decorator from the scriptHandler module is used to assign the NVDA+shift+v shortcut to this script. However, the details of the script and its binding are not important for the purposes of this example. The most important part is the class itself. More information about scripts and the script decorator can be found in the Defining script properties section of this guide.

As with other examples in this guide, remember to delete the created Global Plugin when finished testing and then restart NVDA or reload plugins, so that original functionality is restored.

NVDA Objects

NVDA represents controls and other GUI elements as NVDA Objects. These NVDA Objects contain standardised properties, such as name, role, value, states and description, which allow other parts of NVDA to query or present information about a control in a generalised way. For example, the OK button in a dialog would be represented as an NVDA Object with a name of "OK" and a role of button. Similarly, a checkbox with a label of "I agree" would have a name of "I agree", a role of checkbox, and if currently checked, a state of checked.

As there are many different GUI Toolkits and platform and accessibility APIs, NVDA Objects abstract these differences into a standard form that NVDA can use, regardless of the toolkit or API a particular control is made with. For example, the Ok button just discussed could be a widget in a Java application, an MSAA object, an IAccessible2 object or a UI Automation element.

NVDA Objects have many properties. Some of the most useful are:

There are also a few simplified navigation properties such as simpleParent, simpleNext, simpleFirstChild and simpleLastChild. These are like their respective navigation properties described above, but NVDA filters out useless objects. These properties are used when NVDA's simple review mode is turned on, which is the default. These simple properties may be easier to use, but the real navigation properties more closely reflect the underlying Operating System structure. Also, these may change in future versions of NVDA as improvements are made to simple review, so they should generally be avoided when programmatically locating specific objects.

When developing plugins, most of the time, it is not important what toolkit or API backs an NVDA Object, as the plugin will usually only access standard properties such as name, role and value. However, as plugins become more advanced, it is certainly possible to delve deeper into NVDA Objects to find out toolkit or API specific information if required.

Plugins make use of NVDA Objects in three particular ways:

Just like App Modules and Global Plugins, NVDA Objects can also define events, scripts and gesture bindings.

Scripts and Gesture Bindings

App Modules, Global Plugins and NVDA Objects can define special methods which can be bound to a particular piece of input such as a key press. NVDA refers to these methods as scripts.

A script is a standard Python instance method with a name starting with "script_"; e.g. "script_sayDateTime".

A script method takes two arguments:

As well as the actual script method, some form of gesture binding must be defined, so that NVDA knows what input should execute the script.

A gesture identifier string is a simple string representation of a piece of input. It consists of a two letter character code denoting the source of the input, an optional device in brackets, a colon (:) and one or more names separated by a plus (+) denoting the actual keys or input values.

Some examples of gesture string identifiers are:

Currently, the input sources in NVDA are:

When NVDA receives input, it looks for a matching gesture binding in a particular order. Once a gesture binding is found, the script is executed and no further bindings are used, nor is that particular gesture passed on automatically to the Operating System.

The order for gesture binding lookup is:

Defining script properties

For NVDA 2018.3 and above, the recommended way to set script properties is by means of the so called script decorator. In short, a decorator is a function that modifies the behavior of a particular function or method. The script decorator modifies the script in such a way that it will be properly bound to the desired gestures. Furthermore, it ensures that the script is listed with the description you specify, and that it is categorised under the desired category in the input gestures dialog.

In order for you to use the script decorator, you will have to import it from the scriptHandler module.

from scriptHandler import script

After that, just above your script definition, add the script decorator, providing it the desired arguments. For example:

--- start ---
        description=_("Speaks the date and time"),
        gestures=["kb:NVDA+shift+t", "kb:NVDA+alt+r"]
    def script_sayDateTime(self, gesture):

--- end ---

In this example, your script will be listed in the input gestures dialog under the "Miscellaneous" category. It will have the description "Speaks the date and time", and will be bound to the "NVDA+shift+t" and "NVDA+alt+r" key combinations on the keyboard.

The following keyword arguments can be used when applying the script decorator:

Though the script decorator makes the script definition process a lot easier, there are more ways of binding gestures and setting script properties. For example, a special "__gestures" Python dictionary can be defined as a class variable on an App Module, Global Plugin or NVDA Object. This dictionary should contain gesture identifier strings pointing to the name of the requested script, without the "script_" prefix. You can also specify a description of the script in the method's "doc" attribute. However, beware not to include an inline docstring at the start of the method if you do not set the "doc" attribute, as it would render the description not translatable. The script decorator does not suffer from this limitation, so you are encouraged to provide inline docstrings as needed when using it. Furthermore, an alternative way of specifying the script's category is by means of setting a "category" attribute on the script method to a string containing the name of the category.

Example 4: A Global Plugin to Find out Window Class and Control ID

The following Global Plugin allows you to press NVDA+leftArrow to have the window class of the current focus announced, and NVDA+rightArrow to have the window control ID of the current focus announced. This example shows you how to define one or more scripts and gesture bindings on a class such as an App Module, Global Plugin or NVDA Object.

Copy and paste the code between (but not including) the start and end markers into a new text file with a name of example3.py, which should be saved in the globalPlugins subdirectory. Be very careful to keep all tabs and spaces intact.

Once saved in the right place, either restart NVDA or choose Reload Plugins found under Tools in the NVDA menu.

--- start ---
#Window utility scripts for NVDA
#Developer guide example 4

import globalPluginHandler
from scriptHandler import script
import ui
import api

class GlobalPlugin(globalPluginHandler.GlobalPlugin):

        description=_("Announces the window class name of the current focus object"),
    def script_announceWindowClassName(self, gesture):
        focusObj = api.getFocusObject()
        name = focusObj.name
        windowClassName = focusObj.windowClassName
        ui.message("class for %s window: %s" % (name, windowClassName))

        description=_("Announces the window control ID of the current focus object"),
    def script_announceWindowControlID(self, gesture):
        focusObj = api.getFocusObject()
        name = focusObj.name
        windowControlID = focusObj.windowControlID
        ui.message("Control ID for %s window: %d" % (name, windowControlID))

--- end ---


When NVDA detects particular toolkit, API or Operating System events, it abstracts these and fires its own internal events on plugins and NVDA Objects.

Although most events are related to a specific NVDA Object (e.g. name change, gain focus, state change, etc.), these events can be handled at various levels. When an event is handled, it is stopped from going further down the chain. However, code inside the event can choose to propagate it further if needed.

The order of levels through which the event passes until an event method is found is:

Events are Python instance methods, with a name starting with "event_" followed by the actual name of the event (e.g. gainFocus).

These event methods take slightly different arguments depending at what level they are defined.

If an event for an NVDA Object is defined on an NVDA Object itself, the method only takes one mandatory argument which is the 'self' argument; i.e. the NVDA Object instance). Some events may take extra arguments, though this is quite rare.

If an event for an NVDA Object is defined on a Global Plugin, App Module or Tree Interceptor, the event takes the following arguments:

Some common NVDA Object events are:

There are many other events, though those listed above are usually the most useful.

For an example of an event handled by an App Module, please refer to example 1 (focus beeps in notepad).

the App Module SleepMode variable

App Modules have one very useful property called "sleepMode", which if set to true almost completely disables NVDA within that application. Sleep Mode is very useful for self voicing applications that have their own screen reading functionality, or perhaps even some games that need full use of the keyboard.

Although sleep mode can be toggled on and off by the user with the key command NVDA+shift+s, a developer can choose to have sleep mode enabled by default for an application. This is done by providing an App Module for that application which simply sets sleepMode to True in the AppModule class.

Example 5: A Sleep Mode App Module

The following code can be copied and pasted in to a text file, then saved in the appModules directory with the name of the application you wish to enable sleep mode for. As always, the file must have a .py extension.

--- start ---
import appModuleHandler

class AppModule(appModuleHandler.AppModule):

    sleepMode = True

--- end ---

Providing Custom NVDA Object Classes

Providing custom NVDA Object classes is probably the most powerful and useful way to improve the experience of an application in an NVDA plugin. This method allows you to place all the needed logic for a particular control altogether in one NVDA Object class for that control, rather than scattering code for many controls across a plugin's events.

There are two steps to providing a custom NVDA Object class:

When defining a custom NVDAObject class, you have many NVDAObject base classes to choose from. These base classes contain the base support for the particular accessibility or OS API underlying the control, such as win32, MSAA or Java access Bridge. You should usually inherit your custom NVDAObject class from the highest base class you need in order to choose your class in the first place. For example, if you choose to use your custom NVDAObject class when the window class name is "Edit" and the window control ID is 15, you should probably inherit from NVDAObjects.window.Window, as clearly you are aware that this is a Window object. Similarly, if you match on MSAA's accRole property, you would probably need to inherit from NVDAObjects.IAccessible.IAccessible. You should also consider what properties you are going to override on the custom NVDA Object. For instance, if you are going to override an IAccessible specific property, such as shouldAllowIAccessibleFocusEvent, then you need to inherit from NVDAObjects.IAccessible.IAccessible.

The chooseNVDAObjectOverlayClasses method can be implemented on app modules or global plugin classes. It takes 3 arguments:

  1. self: the app module or global plugin instance.
  2. obj: the NVDAObject for which classes are being chosen.
  3. clsList: a Python list of NVDAObject classes that will be used for obj.

Inside this method, you should decide which custom NVDA Object class(es) (if any) this NVDA Object should use by checking its properties, etc. If a custom class should be used, it must be inserted into the class list, usually at the beginning. You can also remove classes chosen by NVDA from the class list, although this is rarely required.

Example 6: Command to Retrieve the Length of Text in an Edit Field Using a Custom NVDA Object

This app module for notepad provides a command to report the number of characters in edit fields. You can activate it using NVDA+l. Notice that the command is specific to edit fields; i.e. it only works while you are focused in an edit field, rather than anywhere in the application.

The following code can be copied and pasted in to a text file, then saved in the appModules directory with the name of notepad.py.

--- start ---
import appModuleHandler
from scriptHandler import script
from NVDAObjects.IAccessible import IAccessible
import controlTypes
import ui

class AppModule(appModuleHandler.AppModule):

    def chooseNVDAObjectOverlayClasses(self, obj, clsList):
        if obj.windowClassName == "Edit" and obj.role == controlTypes.Role.EDITABLETEXT:
            clsList.insert(0, EnhancedEditField)

class EnhancedEditField(IAccessible):

    def script_reportLength(self, gesture):
        ui.message("%d" % len(self.value))

--- end ---

Making Small Changes to an NVDA Object in App Modules

Sometimes, you may wish to make only small changes to an NVDA Object in an application, such as overriding its name or role. In these cases, you don't need the full power of a custom NVDA Object class. To do this, you can use the NVDAObject_init event available only on App Modules.

The event_NVDAObject_init method takes two arguments:

  1. self: the AppModule instance.
  2. obj: the NVDAObject being initialized.

Inside this method, you can check whether this object is relevant and then override properties accordingly.

Example 7: Labelling the Notepad Edit Field Using event_NVDAObject_init

This app module for notepad makes NVDA report Notepad's main edit field as having a name of "content". That is, when it receives focus, NVDA will say "Content edit".

The following code can be copied and pasted in to a text file, then saved in the appModules directory with the name of notepad.py.

--- start ---
import appModuleHandler
from NVDAObjects.window import Window

class AppModule(appModuleHandler.AppModule):

    def event_NVDAObject_init(self, obj):
        if isinstance(obj, Window) and obj.windowClassName == "Edit" and obj.windowControlID == 15:
            obj.name = "Content"
--- end ---

Parsing additional command line arguments in your plugin

By default NVDA accepts limited set of command line arguments and shows an error for an unknown ones. However if you want to use any additional arguments, this is possible by adding a handler to the extension point addonHandler.isCLIParamKnown. Note that since command line arguments are processed just after NVDA starts, your add-on needs to process them in the global plugin, since app modules or other drivers may not be loaded at this stage. A sample handler can be written as follows:

def processArgs(cliArgument: str) -> bool:
    if cliArgument == "--enable-addon-feature":
        return True  # Argument is known to the add-on and should not be flagged by NVDA
    return False  # unknown argument - NVDA should warn user

Then the handler needs to be registered - preferably in the constructor of your global plugin:


Packaging Code as NVDA Add-ons

To make it easy for users to share and install plugins and drivers, they can be packaged in to a single NVDA add-on package which the user can then install into a copy of NVDA via the Add-on Store found under Tools in the NVDA menu. Add-on packages are only supported in NVDA 2012.2 and later. An add-on package is simply a standard zip archive with the file extension of "nvda-addon" which contains a manifest file, optional install/uninstall code and one or more directories containing plugins and/or drivers.

Non-ASCII File Names in Zip Archives

If your add-on includes files which contain non-ASCII (non-English) characters, you should create the zip archive such that it uses UTF-8 file names. This means that these files can be extracted properly on all systems, regardless of the system's configured language. Unfortunately, many zip archivers do not support this, including Windows Explorer. Generally, it has to be explicitly enabled even in archivers that do support it. 7-Zip supports this, though it must be enabled by specifying the "cu=on" method parameter.

Manifest Files

Each add-on package must contain a manifest file named manifest.ini. This must be a UTF-8 encoded text file. This manifest file contains key = value pairs declaring info such as the add-on's name, version and description.

Available Fields

Although it is highly suggested that manifests contain all fields, the fields marked as mandatory must be included. Otherwise, the add-on will not install.

All string values must be enclosed in quotes as shown in the example below.

The lastTestedNVDAVersion field in particular is used to ensure that users can be confident about installing an add-on. It allows the add-on author to make an assurance that the add-on will not cause instability, or break the users system. When this is not provided, or is less than the current version of NVDA (ignoring minor point updates e.g. 2018.3.1) then the user will be warned not to install the add-on.

An Example Manifest File

--- start ---
name = "myTestAddon"
summary = "Cool Test Add-on"
version = "1.0.0"
description = "An example add-on showing how to create add-ons!"
author = "Michael Curran <mick@example.com>"
url = "https://github.com/nvaccess/nvda/blob/master/projectDocs/dev/addons.md"
docFileName = "readme.html"
minimumNVDAVersion = "2021.1"
lastTestedNVDAVersion = "2022.3.3"
--- end ---

Plugins and Drivers

The following plugins and drivers can be included in an add-on:

Optional install / Uninstall code

If you need to execute code as your add-on is being installed or uninstalled from NVDA (e.g. to validate license information or to copy files to a custom location), you can provide a Python file called installTasks.py in the archive which contains special functions that NVDA will call while installing or uninstalling your add-on. This file should avoid loading any modules that are not absolutely necessary, especially Python C extensions or dlls from your own add-on, as this could cause later removal of the add-on to fail. However, if this does happen, the add-on directory will be renamed and then deleted after the next restart of NVDA. Finally, it should not depend on the existence or state of other add-ons, as they may not be installed, have already been removed or not yet be initialized.

the onInstall function

NVDA will look for and execute an onInstall function in installTasks.py after it has finished extracting the add-on into NVDA. Note that although the add-on will have been extracted at this time, its directory will have a .pendingInstall suffix until NVDA is restarted, the directory is renamed and the add-on is really loaded for the first time. If this function raises an exception, the installation of the add-on will fail and its directory will be cleaned up.

The onUninstall Function

NVDA will look for and execute an onUninstall function in installTasks.py when NVDA is restarted after the user has chosen to remove the add-on. After this function completes, the add-on's directory will automatically be removed. As this happens on NVDA startup before other components are initialized, this function cannot request input from the user.

Localizing Add-ons

It is possible to provide locale-specific information and messages for your add-on. Locale information can be stored in a locale directory in the archive. This directory should contain directories for each language it supports, using the same language code format as the rest of NVDA; e.g. en for English, fr_CA for French Canadian.

Locale-specific Manifest Files

Each of these language directories can contain a locale-specific manifest file called manifest.ini, which can contain a small subset of the manifest fields for translation. These fields are summary and description. All other fields will be ignored.

Locale-specific Messages

Each language directory can also contain gettext information, which is the system used to translate the rest of NVDA's user interface and reported messages. As with the rest of NVDA, an nvda.mo compiled gettext database file should be placed in the LC_MESSAGES directory within this directory. To allow plugins in your add-on to access gettext message information via calls to _(), ngettext(), npgettext() and pgettext() you must initialize translations at the top of each Python module by calling addonHandler.initTranslation(). This function cannot be called in modules that do not belong to an add-on, e.g. in a scratchpad subdirectory. For more information about gettext and NVDA translation in general, please read the Translating NVDA wiki page

Add-on Documentation

Documentation for an add-on should be placed in a doc directory in the archive. Similar to the locale directory, this directory should contain directories for each language in which documentation is available.

Users can access documentation for a particular add-on by opening the Add-on Store, selecting the add-on and pressing the Add-on help button. This will open the file named in the docFileName parameter of the manifest. NVDA will search for this file in the appropriate language directories. For example, if docFileName is set to readme.html and the user is using English, NVDA will open doc\en\readme.html.

NVDA Python Console

The NVDA Python console emulates the interactive Python interpreter from within NVDA. It is a development tool which is useful for debugging, general inspection of NVDA internals or inspection of the accessibility hierarchy of an application.


The console can be activated in two ways:

The console is similar to the standard interactive Python interpreter. Input is accepted one line at a time and processed when enter is pressed. Multiple lines can be pasted at once from the clipboard and will be processed one by one. You can navigate through the history of previously entered lines using the up and down arrow keys.

Output (responses from the interpreter) will be spoken when enter is pressed. The f6 key toggles between the input and output controls. When on the output control, alt+up/down jumps to the previous/next result (add shift for selecting). Pressing control+l clears the output.

The result of the last executed command is stored in the "" global variable. This shadows the gettext function which is stored as a built-in with the same name. It can be unshadowed by executing "del " and avoided altogether by executing " = ".

Closing the console window (with escape or alt+F4) simply hides it. This allows the user to return to the session as it was left when it was closed, including history and variables.


Automatic Imports

For convenience, the following modules and variables are automatically imported in the console: sys, os, wx, log (from logHandler), api, queueHandler, config, controlTypes, textInfos, braille, speech, vision, appModules, globalPlugins

See: pythonConsole.PythonConsole.initNamespace

Snapshot Variables

Whenever NVDA+control+z is pressed, certain variables available in the console will be assigned according to the current state of NVDA. These variables are:

Tab completion

The input control supports tab-completion of variables and member attributes names. Hit the tab key once to complete the current input if there is one single candidate. If there is more than one, hit the tab key a second time to open a menu listing all matching possibilities. By default, only "public" member attributes are listed. That is, if the input is "nav.", attribute names with no leading underscore are proposed. If the input is "nav._", attribute names with a single leading underscore are proposed. Similarly, if the input is "nav.__", attribute names with two leading underscores are proposed.

Remote Python Console

A remote Python console is available in source builds of NVDA, for situations where remote debugging of NVDA is useful. It is similar to the local Python console discussed above, but is accessed via TCP.

Please be aware that this is a huge security risk. It is not available in binary builds distributed by NV Access, and You should only enable it if you are connected to trusted networks.


To enable the remote Python console, use the local Python console to import remotePythonConsole and call remotePythonConsole.initialize(). You can then connect to it via TCP port 6832.

History of previously entered lines is not supported.

The namespace is the same as the namespace in the local Python console.

There are some special functions:

Extension Points

NVDA's extensionPoints module, allows code in different parts of NVDA, or in add-ons, to perform tasks such as:

There are five kinds of extension point:

Type Purpose
Action Allows some code to find out what other code is doing. For example, an add-on can be notified before or after a config profile changes.
Filter Edits data. A filter registered in the speech module, might allow changing speech strings before they are spoken.
Decider Runs each registered handler until one of them returns False. If one does, it can be used to prevent the invoking code from running.
AccumulatingDecider Like Decider, but always runs all of its registered handlers, and only decides if one of them failed at the end. The expected result of each is True by default, though expecting False is possible.
Chain Allows registering handlers that return iterables (mainly generators). Calling iter on the Chain returns a generator that iterates over all the handlers.

The sections below provide the list of currently defined extension points in NVDA, along with brief descriptions for them. Please see code documentation in the associated files, or the code itself, for further explanation. The section titles below represent the package or module in which the listed extension points are defined.

For examples of how to define and use new extension points, please see the code documentation of the extensionPoints package.


Type Extension Point Description
Filter filter_displaySize Allows components or add-ons to change the display size used for braille output.
Action displaySizeChanged Notifies of display size changes.
Action pre_writeCells Notifies when cells are about to be written to a braille display
Action displayChanged Notifies of braille display changes.
Decider decide_enabled Allows deciding whether the braille handler should be forcefully disabled.


Type Extension Point Description
Action post_appSwitch Triggered when the foreground application changes


Type Extension Point Description
AccumulatingDecider isCLIParamKnown Allows adding NVDA commandline parameters which apply to plugins. See this section of the Dev Guide for more information.


Type Extension Point Description
Action postBrailleViewerToolToggledAction Triggered every time the Braille Viewer is created / shown or hidden / destroyed.


Type Extension Point Description
Action post_configProfileSwitch Notifies after the configuration profile has been switched.
Action pre_configSave Notifies before NVDA's configuration is saved to disk.
Action post_configSave Notifies after NVDA's configuration has been saved to disk.
Action pre_configReset Notifies before configuration is reloaded from disk or factory defaults are applied.
Action post_configReset Notifies after configuration has been reloaded from disk or factory defaults were applied.


Type Extension Point Description
Action postNvdaStartup Notifies after NVDA has finished starting up.


Type Extension Point Description
Decider decide_executeGesture Notifies when a gesture is about to be executed, allowing other code to decide if it should be.


Type Extension Point Description
Action _onErrorSoundRequested Triggered every time an error sound needs to be played. This extension point should not be used directly but retrieved calling getOnErrorSoundRequested() instead.


Type Extension Point Description
Decider decide_playWaveFile Notifies when a wave file is about to be played, allowing other code to decide if it should be.


Type Extension Point Description
Action speechCanceled Triggered when speech is canceled.
Action pre_speechCanceled Triggered before speech is canceled.
Action pre_speech Triggered before NVDA handles prepared speech.
Filter filter_speechSequence Allows components or add-ons to filter speech sequence before it passes to the synth driver.


Type Extension Point Description
Action synthIndexReached Notifies when a synthesizer reaches an index during speech.
Action synthDoneSpeaking Notifies when a synthesizer finishes speaking.
Action synthChanged Notifies of synthesizer changes.


Type Extension Point Description
Decider decide_beep Notifies when a beep is about to be generated and played, allowing a component to decide whether it should be.


Type Extension Point Description
Action post_browseModeStateChange Notifies when browse mode state changes.


Type Extension Point Description
Action post_sessionLockStateChanged Notifies when a session lock or unlock event occurs.


Type Extension Point Description
Action pre_handleWindowMessage Notifies when NVDA receives a window message, allowing components to perform an action when certain system events occur.


Type Extension Point Description
Action winAPI.secureDesktop.post_secureDesktopStateChange Notifies when the user has switched to/from the secure desktop


Type Extension Point Description
Chain scanForDevices Can be iterated to scan for braille devices.


These extension points are expected to be used and registered to differently than other extension points. Please see the EventExtensionPoints class documentation for more information, and detailed descriptions.

Type Extension Point Notifies a vision enhancement provider when ...
Action post_objectUpdate an object property has changed.
Action post_focusChange the focused NVDAObject has changed.
Action post_foregroundChange the foreground NVDAObject has changed.
Action post_caretMove a physical caret has moved.
Action post_browseModeMove a virtual caret has moved.
Action post_reviewMove the position of the review cursor has changed.
Action post_mouseMove the mouse has moved.
Action post_coreCycle the end of each core cycle has been reached.