Plugins¶
Using plugins¶
File-based plugins¶
Place the plugin file in either:
The main
plugins
directory of this moduleA
plugins
directory in your profile
Use the /config files
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 setting plugins.enabled
will overwrite the default enabled plugins. Use the /plugins
command for a list of currently enabled plugins.
Core plugins¶
These plugins are built into LWE core:
echo: Simple echo plugin, echos back the text you give it
examples: Easily install example configuration files (see Installing examples)
They can be disabled by removing them from plugins.enabled
in your configuration file.
LWE maintained plugins¶
These plugins are maintained by the LWE team, and are not part of LWE core – they must be installed.
Instructions for installing and configuring each plugin can be found at the referenced repository for each plugin.
Shell command plugins¶
These plugins add additional commands to the shell:
- awesome: Use prompts from Awesome ChatGPT Prompts
- database: Send natural language commands to a database WARNING: POTENTIALLY DANGEROUS – DATA INTEGRITY CANNOT BE GUARANTEED.
- 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 to https://gist.github.com
- pastebin: Post a conversation to https://pastebin.com
- shell: Transform natural language into a shell command, and optionally execute it WARNING: POTENTIALLY DANGEROUS – YOU ARE RESPONSIBLE FOR VALIDATING THE COMMAND RETURNED BY THE LLM, AND THE OUTCOME OF ITS EXECUTION.
- test: Test plugin, echos back the command you give it
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 storage mechanism the chat-based providers use.
Supported providers¶
NOTE: While these provider integrations are working, none have been well-tested yet.
- provider_ai21: Access to AI21 models
https://github.com/llm-workflow-engine/lwe-plugin-provider-ai21
- provider_azure_openai_chat: Access to Azure OpenAI chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-azure-openai-chat
- provider_chat_anthropic: Access to Anthropic chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-anthropic
- provider_chat_anyscale: Access to Anyscale chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-anyscale
- provider_chat_cohere: Access to Cohere chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-cohere
- provider_chat_fireworks: Access to Fireworks chat models.
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-fireworks
- provider_chat_google_genai: Access to Google GenAI chat models.
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-google-genai
- provider_chat_groq: Access to Groq chat models.
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-groq
- provider_chat_mistralai: Access to MistralAI chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-mistralai
- provider_chat_ollama: Access to Ollama chat models
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-ollama
- provider_chat_vertexai: Access to Google Vertex AI chat models.
https://github.com/llm-workflow-engine/lwe-plugin-provider-chat-vertexai
- provider_cohere: Access to Cohere models
https://github.com/llm-workflow-engine/lwe-plugin-provider-cohere
- provider_huggingface_hub: Access to Hugging Face Hub models
https://github.com/llm-workflow-engine/lwe-plugin-provider-huggingface-hub
- provider_openai: Access to non-chat OpenAI models (GPT-3, etc.)
https://github.com/llm-workflow-engine/lwe-plugin-provider-openai
- provider_openrouter: Access to OpenRouter models
https://github.com/llm-workflow-engine/lwe-plugin-provider-openrouter
- provider_vertexai: Access to Google Vertex AI text/code models.
https://github.com/llm-workflow-engine/lwe-plugin-provider-vertexai
Usage¶
Use the /providers
command for a list of currently enabled providers.
See /help provider
for how to switch providers/models on the fly.
Example:
/provider openai
/model model_name text-davinci-003
Writing plugins¶
There is currently no developer documentation for writing plugins.
The plugins
directory has some default plugins, examining those will give a good idea for how to design a new one.
In particular, the echo
plugin is well commented. The package plugins listed above also contain many different
approaches you can learn from.
To write new provider plugins, investigate the existing provider plugins as examples.
Currently, plugins for the shell can only add new commands.
Plugin structure¶
In order for plugins to load, a few simple conventions must be followed:
All plugins must inherit from the base Plugin class, and provide implementations of the
setup()
anddefault_config()
methods. Class name should be a camel-cased version of the plugin name:from lwe.core.plugin import Plugin class ExamplePlugin(Plugin): """ An example plugin, does blah blah blah... """ # Implement these... @abstractmethod def setup(self): pass @abstractmethod def default_config(self): pass
The first line of the class docstring will be used as the plugin description.
- Naming conventions: Consider a plugin named
example_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 be
lwe_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" ] }, )
- Naming conventions: Consider a plugin named
Available objects¶
An instantiated plugin has access to these objects.
self.config
: The current instantiated Config objectself.log
: The instantiated Logger objectself.backend
: The instantiated backendself.shell
: The instantiated shell