How to output structured JSON from DuckDuckGo tool call

[ajesuscode]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

import dotenv from 'dotenv';
import { z } from 'zod';
import { DuckDuckGoSearch } from "@langchain/community/tools/duckduckgo_search";
import { ChatPromptTemplate } from "@langchain/core/prompts";
import { RunnableLambda } from "@langchain/core/runnables";
import { HumanMessage } from "@langchain/core/messages";
import { ChatOpenAI } from "@langchain/openai";


dotenv.config();

const resultsSchema = z.object({
  results: z.array(
    z.object({
      websiteName: z.string(),
      price: z.string(),
      link: z.string(),
    })
  ),
});

const llm = new ChatOpenAI({
  model: "gpt-4",
  temperature: 0,
})

const tool = new DuckDuckGoSearch({ maxResults: 10 });

const prompt = ChatPromptTemplate.fromMessages([
    ["system", "Your task is to list all prices in format Website Name - Price - link. Nothing else"],
    ["placeholder", "{messages}"],
  ]);

  const llmWithTools = llm.bindTools([tool]);
  
  const chain = prompt.pipe(llmWithTools).pipe(jsonOutput);
  
  const toolChain = RunnableLambda.from(async (userInput, config) => {
    const humanMessage = new HumanMessage(userInput);
    const aiMsg = await chain.invoke(
      {
        messages: [new HumanMessage(userInput)],
      },
      config
    );
    const toolMsgs = await tool.batch(aiMsg.tool_calls, config);
    console.log("TOOL",toolMsgs);
    return chain.with.invoke(
      {
        messages: [humanMessage, aiMsg, ...toolMsgs],
      },
      config
    );
  });

  const postReqMessage = "Fender Straticaster";
  
  const toolChainResult = await toolChain.invoke(
    `${postReqMessage} list all current prices`
  );
  console.log("TOOLCHAIN RESULT", toolChainResult);
  const { content } = toolChainResult;

  console.log(
    "AIMessage",
    JSON.stringify(
      {
        tool_calls,
        content,
      },
      null,
      2
    )
  );

Description

I am using an example from official documentation here: duckduckgo_search
The output is a string, but I want it to be a structured JSON
When trying to add method withStructuredJson()
I got error, that basically says that bindTools method doesnt exist on the object
Who knows how to achieve a structured output after calling DuckDuckGo tool
Please help, or point to a working similar solution
Thank you

System Info

{
"type": "module",
"dependencies": {
"@langchain/community": "^0.3.24",
"@langchain/core": "^0.3.30",
"@langchain/openai": "^0.3.17",
"dotenv": "^16.4.7",
"duck-duck-scrape": "^2.2.7",
"langchain": "^0.3.11",
"zod": "^3.24.1"
}
}

[adocherty]

Hi, It would help if you could give the code including the withStructuredJson call that actually produces the error and a clearer idea of what you want the code to do.

From what I understand, making a tool call will return structured data which is passed back to the calling program to then execute the tool call.

Therefore you'll need at least two LLM calls, the first LLM has the tool(s) bound to it. Then call the tools with the returned arguments. Next call the LLM again with a structured output request (see https://python.langchain.com/docs/concepts/structured_outputs/)

However, I'm not sure here why you need the duckduckgo scraping as a tool call here?