Agents are systems that use large language models (LLMs) to decide on actions in a repeated manner, leveraging external sources of knowledge or computation. This approach offers benefits such as combining LLMs with tools, iterative planning and action taking, and robust error handling. Toolkits allow developers to group and initialize tools sharing a particular resource, enabling the construction of agents for specific use-cases. Examples include SQLDatabaseAgent, OpenAPI Agent, JSON agent, Vectorstore agent, Python agent, Pandas DataFrame agent, and CSV agent, each leveraging their respective toolkits to achieve unique capabilities.