DOC: inaccurate description in the structured output conceptual guide

[saharis9988]

Checklist

• I added a very descriptive title to this issue.
• I included a link to the documentation page I am referring to (if applicable).

Issue with current documentation:

Hi team,
I’m working on understanding how structured output is implemented without relying on the withStructuredOutput method, in order to better grasp the underlying process.
In my attempt to implement this via tool calling, I ran into an issue when trying to bind a Zod schema directly to bindTools (following the documentation). An exception is thrown because the schema doesn’t implement the necessary tool interface.
It seems the documentation example should explicitly show how to create a tool from the schema before passing it to bindTools, as this step appears to be missing..
docs

Idea or request for content:

No response

[dosubot]

Hey @saharis9988! I'm here to help you with any bugs, questions, or contributions you have. Let's work on resolving this issue together.

To address the issue you're encountering, you can create a tool from a Zod schema using the tool function from the @langchain/core/tools package before passing it to bindTools. Here's an example of how to do this:

import { z } from "zod";
import { tool } from "@langchain/core/tools";

const myToolSchema = z.object({
  param1: z.string().describe("Description for param1"),
  param2: z.number().describe("Description for param2"),
});

const myTool = tool(async (input): Promise<string> => {
  // Your custom logic here
  return `Processed ${input.param1} and ${input.param2}`;
}, {
  name: "my_custom_tool",
  description: "A custom tool that processes param1 and param2",
  schema: myToolSchema,
});

// Example invocation
await myTool.invoke({ param1: "example", param2: 42 });

Once you have defined your tool, you can bind it to a chat model using the .bindTools() method:

import { ChatGoogleGenerativeAI } from "@langchain/google-genai";

const llm = new ChatGoogleGenerativeAI({
  model: "gemini-1.5-pro",
  temperature: 0,
  maxRetries: 2,
  // other params...
});

const llmWithTools = llm.bindTools([myTool]);

// Now you can invoke the model with the tool bound
const res = await llmWithTools.invoke("Your prompt here");
console.log(res);

This setup allows the chat model to generate arguments for the tool, and you can then decide whether to execute the tool based on these arguments ^[1][2][3].

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}