Parallel tool calls

Parallel tool calls are handled by ToolNode: it reads the tool calls in the latest AIMessage, runs the requested tools concurrently, and returns a separate ToolMessage for each call. This is useful in a StateGraph or MessageGraph when independent operations can be performed during the same model step.

new ToolNode(tools)

A list of ToolMessages, one for each tool call.

import { ToolNode } from "@langchain/langgraph/prebuilt";
import { AIMessage, ToolMessage } from "@langchain/core/messages";

const tools = [
  async function getWeather({ city }: { city: string }) {
    return `${city}: sunny`;
  },
  async function getTime({ city }: { city: string }) {
    return `${city}: 12:00`;
  },
];

const toolNode = new ToolNode(tools);
const message = new AIMessage({
  content: "",
  tool_calls: [
    { name: "getWeather", args: { city: "Boston" }, id: "1", type: "tool_call" },
    { name: "getTime", args: { city: "Tokyo" }, id: "2", type: "tool_call" },
  ],
});

const result = await toolNode.invoke({ messages: [message] });
console.log(result.messages.map((m: ToolMessage) => m.content));

ToolNode receives one AI message containing multiple tool calls and executes those calls in parallel, returning one ToolMessage for each result.