File-based plugins¶

Place the plugin file in either:

• The mainplugins directory of this module

• Aplugins directory in your profile

Use the/configfiles command to see a list of currently configured plugin paths.

Package-based plugins¶

Install the plugin package – see below for a list of package plugins.

Enabling plugins¶

Enable the plugin in your configuration:

plugins:enabled:# This is a list of plugins to enable, each list item# should be the name of a plugin file.# For file-based plugins, this is the filename without the# extension.# For package-based plugins, see the installation instructions# for the package.-test

Note that settingplugins.enabled will overwrite the default enabled plugins. Use the/plugins command for a list of currently enabled plugins.

Reloading plugins¶

To dynamically reload a plugin:

/pluginreload[plugin_name]

Reloading a plugin will also clear the main cache file for the plugin – as some provider plugins cache data on current models, this command can be used to refresh the available models for the provider. For example:

/pluginreloadprovider_chat_fireworks

Shell command plugins¶

These plugins add additional commands to the shell:

• awesome: Use prompts fromAwesome ChatGPT Prompts

  https://github.com/llm-workflow-engine/lwe-plugin-awesome

• database: Send natural language commands to a databaseWARNING: POTENTIALLY DANGEROUS – DATA INTEGRITY CANNOT BE GUARANTEED.

  https://github.com/llm-workflow-engine/lwe-plugin-database

• data_query: Send natural language commands to a loaded file of structured data

  https://github.com/llm-workflow-engine/lwe-plugin-data-query

• gist: Post a conversation tohttps://gist.github.com

  https://github.com/llm-workflow-engine/lwe-plugin-gist

• pastebin: Post a conversation tohttps://pastebin.com

  https://github.com/llm-workflow-engine/lwe-plugin-pastebin

• shell: Transform natural language into a shell command, and optionally execute itWARNING: POTENTIALLY DANGEROUS – YOU ARE RESPONSIBLE FOR VALIDATING THE COMMAND RETURNED BY THE LLM, AND THE OUTCOME OF ITS EXECUTION.

  https://github.com/llm-workflow-engine/lwe-plugin-shell

• test: Test plugin, echos back the command you give it

  https://github.com/llm-workflow-engine/lwe-plugin-test

Provider plugins¶

These plugins add additional LLM providers:

NOTE: Most provider plugins are NOT chat-based, and instead return a single response to any text input.These inputs and responses are still managed as ‘conversations’ for storage purposes, using the same storagemechanism the chat-based providers use.

/provider openai/model model_name text-davinci-003

Plugin structure¶

In order for plugins to load, a few simple conventions must be followed:

1. All plugins must inherit from the basePlugin class,and provide implementations of thesetup() anddefault_config() methods.Class name should be a camel-cased version of the plugin name:

   fromlwe.core.pluginimportPluginclassExamplePlugin(Plugin):"""    An example plugin, does blah blah blah...    """# Implement these...@abstractmethoddefsetup(self):pass@abstractmethoddefdefault_config(self):pass

   The first line of the class docstring will be used as the plugin description.

2. Naming conventions: Consider a plugin namedexample_plugin:

   • File-based plugin: The filename must be the plugin name with a.py extension,example_plugin.py

   • Package-based plugin: The the entry point must belwe_plugins, and the plugin name must be prefixed withlwe-plugin-:

     setup(name="lwe-plugin-example-plugin",# Other setup options...entry_points={"lwe_plugins":["lwe_plugin_example_plugin = lwe_plugin_example_plugin.plugin:ExamplePlugin"]},)

Available objects¶

An instantiated plugin has access to these objects.

• self.config: The current instantiated Config object

• self.log: The instantiated Logger object

• self.backend: The instantiated backend

• self.shell: The instantiated shell