Nvim :help pages, generated from source using the tree-sitter-vimdoc parser.

Nvim support for remote plugins

1. Introduction remote-plugin-intro

Extensibility is a primary goal of Nvim. Any programming language may be used to extend Nvim without changes to Nvim itself. This is achieved with remote plugins, coprocesses that have a direct communication channel (via RPC) with the Nvim process.
Even though these plugins run in separate processes they can call, be called, and receive events just as if the plugin's code were executed in the main process.

2. Plugin hosts remote-plugin-hosts

While plugins can be implemented as arbitrary programs that communicate directly with the high-level Nvim API and are called via rpcrequest() and rpcnotify(), that is not the best approach. Instead, developers should first check whether a plugin host is available for their chosen programming language.
Plugin hosts are programs that provide a high-level environment for plugins, taking care of most boilerplate involved in defining commands, autocmds, and functions implemented over RPC connections. Hosts are loaded only when one of their registered plugins require it, keeping Nvim's startup as fast as possible, even if many plugins/hosts are installed.
The best way to learn about remote plugins is with an example, so let's see what a Python plugin looks like. This plugin exports a command, a function, and an autocmd. The plugin is called 'Limit', and all it does is limit the number of requests made to it. Here's the plugin source code:
import pynvim
class Limit(object):
    def __init__(self, vim):
        self.vim = vim
        self.calls = 0
    @pynvim.command('Cmd', range='', nargs='*', sync=True)
    def command_handler(self, args, range):
        self.vim.current.line = (
            'Command: Called %d times, args: %s, range: %s' % (self.calls,
    @pynvim.autocmd('BufEnter', pattern='*.py', eval='expand("<afile>")',
    def autocmd_handler(self, filename):
        self.vim.current.line = (
            'Autocmd: Called %s times, file: %s' % (self.calls, filename))
    def function_handler(self, args):
        self.vim.current.line = (
            'Function: Called %d times, args: %s' % (self.calls, args))
    def _increment_calls(self):
        if self.calls == 5:
            raise Exception('Too many calls!')
        self.calls += 1
As can be seen, the plugin is implemented using idiomatic Python (classes, methods, and decorators). The translation between these language-specific idioms to Vimscript occurs while the plugin manifest is being generated (see the next section).
Notice that the exported command and autocmd are defined with the "sync" flag, which affects how Nvim calls the plugin: with "sync" the rpcrequest() function is used, which will block Nvim until the handler function returns a value. Without the "sync" flag, the call is made using a fire and forget approach with rpcnotify(), meaning return values or exceptions raised in the handler function are ignored.
To test the above plugin, it must be saved in "rplugin/python" in a 'runtimepath' directory (~/.config/nvim/rplugin/python/ for example). Then, the remote plugin manifest must be generated with :UpdateRemotePlugins.
Just installing remote plugins to "rplugin/{host}" isn't enough for them to be automatically loaded when required. You must execute :UpdateRemotePlugins every time a remote plugin is installed, updated, or deleted.
:UpdateRemotePlugins generates the remote plugin manifest, a special Vimscript file containing declarations for all Vimscript entities (commands/autocommands/functions) defined by all remote plugins, with each entity associated with the host and plugin path.
Manifest declarations are just calls to the remote#host#RegisterPlugin function, which takes care of bootstrapping the host as soon as the declared command, autocommand, or function is used for the first time.
The manifest generation step is necessary to keep Nvim's startup fast in situations where a user has remote plugins with different hosts. For example, say a user has three plugins, for Python, Java and .NET hosts respectively. If we were to load all three plugins at startup, then three language runtimes would also be spawned, which could take seconds!
With the manifest, each host will only be loaded when required. Continuing with the example, say the Java plugin is a semantic completion engine for Java code. If it defines the autocommand BufEnter *.java, then the Java host is spawned only when Nvim loads a buffer matching "*.java".
If the explicit call to :UpdateRemotePlugins seems inconvenient, try to see it like this: It's a way to provide IDE capabilities in Nvim while still keeping it fast and lightweight for general use. It's also analogous to the :helptags command.
$NVIM_RPLUGIN_MANIFEST Unless $NVIM_RPLUGIN_MANIFEST is set the manifest will be written to a file named rplugin.vim at:
$XDG_DATA_HOME/nvim/ or ~/.local/share/nvim/
$LOCALAPPDATA/nvim/ or ~/AppData/Local/nvim/
Commands index
Quick reference