Tavily research integration | LangChain 简体中文

Tavily is a search engine built specifically for AI agents (LLMs), delivering real-time, accurate, and factual results at speed. Tavily offers a Research endpoint that generates comprehensive research reports tailored for LLMs and RAG.

Overview

Integration details

Class	Package	PY support	Downloads	Version
`TavilyResearch`	`@langchain/tavily`	✅

Tool features

Returns artifact	Native async	Return data	Pricing
❌	✅	Research report content, sources, citations	1,000 free credits / month

Setup

The integration lives in the @langchain/tavily package:

npm

yarn

pnpm

npm install @langchain/tavily @langchain/core

yarn add @langchain/tavily @langchain/core

pnpm add @langchain/tavily @langchain/core

Credentials

Set up an API key on the Tavily dashboard and set it as an environment variable named TAVILY_API_KEY.

process.env.TAVILY_API_KEY = "YOUR_API_KEY";

It is also helpful (but not needed) to set up LangSmith for observability:

process.env.LANGSMITH_TRACING = "true";
process.env.LANGSMITH_API_KEY = "your-api-key";

Instantiation

You can import and instantiate TavilyResearch like this:

import { TavilyResearch } from "@langchain/tavily";

const tool = new TavilyResearch({
  // model: "mini",
  // citationFormat: "apa",
  // stream: false,
});

model (optional): Research agent model. "mini", "pro", or "auto" (default).
citationFormat (optional): Citation format for sources. "numbered", "mla", "apa", or "chicago" (default "numbered").
outputSchema (optional): JSON Schema that shapes the research output.
stream (optional): Whether to stream research results. Default is false.

Invocation

Invoke directly with args

The Tavily research tool accepts the following arguments during invocation:

input (required): The research task or question to investigate
Optional overrides: model, outputSchema, stream, citationFormat

await tool.invoke({
  input: "What are the latest developments in AI?",
});

Invoke with ToolCall

You can also invoke the tool with a model-generated ToolCall, in which case a ToolMessage is returned:

// This is usually generated by a model, but we'll create a tool call directly for demo purposes.
const modelGeneratedToolCall = {
  args: {
    input: "What are the latest developments in AI?",
  },
  id: "1",
  name: tool.name,
  type: "tool_call",
};

await tool.invoke(modelGeneratedToolCall);

Use within an agent

Pass the research tool to createAgent so the agent can call it with a research task:

import { ChatOpenAI } from "@langchain/openai";
import { TavilyResearch } from "@langchain/tavily";
import { createAgent } from "langchain";

const llm = new ChatOpenAI({
  model: "gpt-5.5",
});

const tavilyResearchTool = new TavilyResearch({
  model: "mini",
});

const agent = createAgent({
  model: llm,
  tools: [tavilyResearchTool],
});

const userInput =
  "Research the latest developments in AI agents and summarize key trends.";

const stream = await agent.streamEvents(
  { messages: [{ role: "user", content: userInput }] },
  { version: "v3" },
);

await Promise.all([
  (async () => {
    for await (const message of stream.messages) {
      for await (const token of message.text) {
        process.stdout.write(token);
      }
    }
  })(),
  (async () => {
    for await (const call of stream.toolCalls) {
      console.dir({ name: call.name, input: call.input }, { depth: null });
      await call.output;
    }
  })(),
]);

await stream.output;

API reference

For detailed documentation of all Tavily Research API features and configurations, see the API reference: docs.tavily.com/documentation/api-reference/endpoint/research