Combine 2 agent in langchain version 0.3

[irufano]

I have 2 agent there are sql agent and chart generation agent, how to combine both agents into one that interconnected with each other?

• first agent:

## SQL agent
system_message = SystemMessage(content=SQL_PREFIX)
sql_agent_executor = create_react_agent(
    llm, tools, state_modifier=system_message, checkpointer=memory
)

• second agent:

## chart generation agent
system_message = SystemMessage(content=PROMPT_PREFIX)
chart_agent_executor = create_react_agent(llm, tools, state_modifier=system_message)

[dosubot[bot]]

I couldn't find specific examples or documentation in LangChain version 0.3 that demonstrate combining SQL and chart generation agents. However, you can consider using the AgentExecutor to manage and orchestrate the execution of multiple agents. The AgentExecutor allows you to define tools and create agents that can invoke these tools as needed, facilitating the integration of different functionalities.

To combine your SQL agent and chart generation agent, you might want to:

1. Define the tools required for both agents.
2. Create a unified agent that can handle tasks for both SQL queries and chart generation.
3. Use the AgentExecutor to manage the execution flow, ensuring that the output of one agent can be used as input for the other if necessary.

For more advanced orchestration, you might explore using classes like BaseMultiActionAgent or OpenAIMultiFunctionsAgent, which are designed for integrating or orchestrating multiple agents ^[1].

To continue talking to Dosu, mention @dosu.

[irufano]

can an agent that create by create_react_agent has 2 ai outputs?

[dosubot[bot]]

The create_react_agent function in LangChain is designed to create an agent that processes input and returns either an AgentAction or AgentFinish. While the function itself doesn't inherently support multiple AI outputs directly, you can configure the AgentOutputParser to handle multiple outputs. The ToolsAgentOutputParser can parse multiple tool calls within a single message, returning a list of AgentAction objects if multiple tool calls are present ^[1]. This means that while the agent can handle multiple outputs, it requires the output parser to be set up to parse and manage these multiple outputs effectively.

To continue talking to Dosu, mention @dosu.

[irufano]

can you give me example? @dosu

[feijoes]

By interconnected, do you mean the agents communicate with each other or that you want a single agent that performs both tasks? If it’s the latter, you could try using this

[mraguth]

@feijoes is it OpenAI works with AzureOpenAI? I try to configure with azureopenai... but not much luck. Any thoughts?