create_csv_agent#

langchain_cohere.csv_agent.agent.create_csv_agent(llm: BaseLanguageModel, path: str | List[str], extra_tools: List[BaseTool] = [], pandas_kwargs: dict | None = None, prompt: ChatPromptTemplate | None = None, number_of_head_rows: int = 5, verbose: bool = True, return_intermediate_steps: bool = True, temp_path_dir: str | None = None, temp_path_prefix: str | None = 'langchain', temp_path_suffix: str | None = 'csv_agent') → AgentExecutor[source]#

Create csv agent with the specified language model.

Parameters:

  • llm (BaseLanguageModel) – Language model to use for the agent.

  • path (str | List[str]) – A string path, or a list of string paths that can be read in as pandas DataFrames with pd.read_csv().

  • number_of_head_rows (int) – Number of rows to display in the prompt for sample data

  • include_df_in_prompt – Display the DataFrame sample values in the prompt.

  • pandas_kwargs (dict | None) – Named arguments to pass to pd.read_csv().

  • prefix – Prompt prefix string.

  • suffix – Prompt suffix string.

  • prompt (ChatPromptTemplate | None) – Prompt to use for the agent. This takes precedence over the other prompt arguments, such as suffix and prefix.

  • temp_path_dir (str | None) – Temporary directory to store the csv files in for the python repl.

  • delete_temp_path – Whether to delete the temporary directory after the agent is done. This only works if temp_path_dir is not provided.

  • extra_tools (List[BaseTool]) –

  • verbose (bool) –

  • return_intermediate_steps (bool) –

  • temp_path_prefix (str | None) –

  • temp_path_suffix (str | None) –

Returns:

An AgentExecutor with the specified agent_type agent and access to a PythonREPL and any user-provided extra_tools.

Return type:

AgentExecutor

Example

from langchain_cohere import ChatCohere, create_csv_agent

llm = ChatCohere(model="command-r-plus", temperature=0)
agent_executor = create_csv_agent(
    llm,
    "titanic.csv"
)
resp = agent_executor.invoke({"input":"How many people were on the titanic?"})
print(resp.get("output"))
On this page