What are multi-action agents?

A multi-action agent returns a list of AgentAction objects per reasoning step rather than a single action. This enables the agent to call multiple tools simultaneously within a single turn, which is useful when several tool calls are independent and don't need to be serialised.

Multi-action agents implement BaseMultiActionAgent, and their plan() method returns List[AgentAction] instead of a single AgentAction. AgentExecutor detects this and executes all returned actions in parallel before feeding their observations back to the agent.

OpenAI's parallel tool calling feature maps directly to this pattern. When you call ChatOpenAI with tools bound via .bind_tools(), the model can return multiple tool calls in a single response, and AgentExecutor (or LangGraph) runs them concurrently:

from langchain_openai import ChatOpenAI

llm_with_tools = ChatOpenAI(model="gpt-4o").bind_tools([search_tool, calculator_tool])

# Model may respond with both a search call AND a calculator call in one step
response = llm_with_tools.invoke("What is the population of France times 2?")
print(response.tool_calls)  # [{name: 'search', ...}, {name: 'calculator', ...}]

For complex coordination of parallel tool execution with state management, LangGraph is better suited than AgentExecutor, as it provides explicit graph edges for parallel branches.