Tool-augmented chat
A conversational assistant needs to answer questions using external functions such as weather lookup.
AI Agents · Function
createReactAgent
Creates a prebuilt ReAct agent graph from a language model and tools. The agent can invoke tools during a message-driven execution loop before returning a response.
How it works
createReactAgent constructs a prebuilt agent graph from a language model and tools. The graph repeatedly lets the model choose whether to call a tool, executes selected tools, and continues until the model produces a final response.
Syntax
createReactAgent({ llm: model, tools: [tool] })Parameters
| Name | Type | Description |
|---|---|---|
llm | BaseChatModel or compatible language-model runnable | Language model used to reason and select whether to call tools. |
tools | Array of tools | Tools available to the agent during its execution loop. |
Returns
An agent graph runnable that accepts message state and produces the resulting agent state.
Key points
- The documented JavaScript example imports createReactAgent and tool from the langchain package.
- A chat model instance is created and supplied as the agent's language model.
- The returned agent is executed asynchronously through its invoke method.
Common mistakes
- The current documentation presents this as a prebuilt ReAct agent rather than a general-purpose graph constructor.
- The function requires its configuration to be passed as an options object.
Example: Create a tool-using ReAct agent
import { createReactAgent, tool } from "langchain";
import { ChatAnthropic } from "@langchain/anthropic";
import { z } from "zod";
const getWeather = tool(
async ({ location }) => `Weather for ${location}: sunny`,
{ name: "get_weather", schema: z.object({ location: z.string() }) }
);
const agent = createReactAgent({
llm: new ChatAnthropic({ model: "claude-3-7-sonnet-latest" }),
tools: [getWeather],
});
const result = await agent.invoke({
messages: [{ role: "user", content: "What is the weather in SF?" }],
});
console.log(result.messages.at(-1)?.content);Creates an agent with a chat model and weather tool, then invokes it with a user message.
When to use it
Conditional tool use
A model must decide dynamically whether a supplied capability is needed for each request.
Prebuilt agent orchestration
An application wants a ready-made agent loop without manually wiring model and tool nodes.
Interview questions
QWhat does createReactAgent do?
It creates a ReAct agent graph that combines a language model with tools. The returned graph can be invoked with a messages state, and the model may use the supplied tools before answering.
Follow-up: How would you add another capability? Pass another tool in the tools array, ensuring it has a callable implementation and schema suitable for model tool selection.
QWhat are the main configuration options?
The llm option supplies the chat model responsible for deciding whether to respond or call a tool. The tools option supplies the functions available to that model.
Follow-up: What happens if no tool is appropriate? The model can finish with a direct response, so tool execution is not required for every invocation.
QHow is the resulting agent run?
Invoke the returned agent with an object whose messages field contains the conversation, such as a user role and content. The result contains the agent's resulting message state.
Follow-up: How can a conversation continue? Preserve the returned messages and provide them in a later invocation together with the next user message.
Related
toolChatAnthropicagent.invokemessages