Skip to main content

Toolkits

Toolkits are collections of tools designed to be used together for specific tasks. They provide convenience loading methods to initialize the tools with a single function call.

For example, you can load a QA toolkit that bundles together a question answering agent, a search tool, and a summarization tool to enable complex workflows with just:

from langchain.toolkits import load_qa_toolkit

qa_tools = load_qa_toolkit(llm)

This handles initializing each tool with the provided llm and returning them ready to use.

Tools

Tools provide interfaces that agents can use to interact with the world. Tools can be things like:

  • Chains
  • Other agents
  • Search utilities
  • Summarization models
  • Translation models

Some tools like chains and agents require a base LLM to initialize. You can pass in an LLM when loading tools:

from langchain.agents import load_tools

tool_names = [...]
llm = ...
tools = load_tools(tool_names, llm=llm)

This will initialize the tools that need an LLM with the provided one.

Document Loaders

Document loaders are used to load data from different sources as Document objects. A Document combines text content with associated metadata.

For example, there are document loaders for:

  • Loading text files
  • Scraping the text contents of a web page
  • Loading transcripts of YouTube videos

Document loaders expose a load method to load data from a source as Document objects. Many also implement lazy_load to lazily load data into memory only when needed.

Here is an example loading a Markdown file:

from langchain.document_loaders import TextLoader

loader = TextLoader("./index.md")  
docs = loader.load()

This loads the Markdown file as a single Document object.