import Admonition from "@theme/Admonition";
Agents are components that use reasoning to make decisions and take actions, designed to autonomously perform tasks or provide services with some degree of agency. LLM chains can only perform hardcoded sequences of actions, while agents use LLMs to reason through which actions to take, and in which order.
The
AgentInitializer
constructs a zero-shot agent from a language model (LLM) and additional tools.
Parameters:
- LLM: The language model used by the
AgentInitializer
.
- Memory: Enables memory functionality, allowing the agent to recall and use information from previous interactions.
- Tools: The tools available to the agent.
- Agent: Specifies the type of agent to instantiate. Currently supported types include
zero-shot-react-description
, react-docstore
, self-ask-with-search
, conversational-react-description
, and openai-functions
.
The
CSVAgent
interacts with CSV (Comma-Separated Values) files, commonly used to store tabular data. Each row in a CSV file represents a record, and each column represents a field. The CSV agent can read and write CSV files, process data, and perform tasks such as filtering, sorting, and aggregating.
Parameters:
- LLM: The language model used by the
CSVAgent
.
- Path: The file path to the CSV data.
The
JSONAgent
manages JSON (JavaScript Object Notation) data. This agent, like the CSVAgent, uses a language model (LLM) and a toolkit for JSON manipulation. It can explore a JSON blob to extract needed information, list keys, retrieve values, and navigate through the JSON structure.
Parameters:
- LLM: The language model used by the
JSONAgent
.
- Toolkit: The toolkit available to the agent.
The
SQLAgent
interacts with SQL databases, capable of querying, retrieving data, and executing SQL statements. It provides insights into the database structure, including tables and schemas, and can perform operations such as insertions, updates, and deletions.
Parameters:
- LLM: The language model used by the
SQLAgent
.
- Database URI: The connection URI for the SQL database.
The
VectorStoreAgent
operates with a vector store, which is a data structure for storing and querying vector-based data representations. This agent can query the vector store to find information relevant to user inputs.
Parameters:
- LLM: The language model used by the
VectorStoreAgent
.
- Vector Store Info: The
VectorStoreInfo
used by the agent.
The
VectorStoreRouterAgent
is a custom agent that uses a vector store router. It is typically used to retrieve information from multiple vector stores connected through a
VectorStoreRouterToolkit
.
Parameters:
- LLM: The language model used by the
VectorStoreRouterAgent
.
- Vector Store Router Toolkit: The toolkit used by the agent.
The
ZeroShotAgent
uses the ReAct framework to decide which tool to use based on the tool's description. It is the most general-purpose action agent, capable of determining the necessary actions and their sequence through an
LLMChain
.
Parameters:
- Allowed Tools: The tools accessible to the agent.
- LLM Chain: The LLM Chain used by the agent.