How to convert tools to OpenAI Functions

This notebook goes over how to use LangChain tools as OpenAI functions.

npm

npm i @langchain/community @langchain/openai

yarn

yarn add @langchain/community @langchain/openai

pnpm

pnpm add @langchain/community @langchain/openai

import { Calculator } from "@langchain/community/tools/calculator";
import { HumanMessage } from "@langchain/core/messages";
import { convertToOpenAIFunction } from "@langchain/core/utils/function_calling";
import { ChatOpenAI } from "@langchain/openai";

const model = new ChatOpenAI({
  model: "gpt-3.5-turbo",
});

const tools = [new Calculator()];
const functions = [convertToOpenAIFunction(tools[0])];

functions[0];

{
  name: "calculator",
  description: "Useful for getting the result of a math expression. The input to this tool should be a valid mathema"... 63 more characters,
  parameters: {
    type: "object",
    properties: { input: { type: "string" } },
    additionalProperties: false,
    "$schema": "http://json-schema.org/draft-07/schema#"
  }
}

const modelWithTools = model.bind({
  functions,
});

const message = await modelWithTools.invoke([
  new HumanMessage("use a calculator to add 2 and 3"),
]);

message;

AIMessage {
  lc_serializable: true,
  lc_kwargs: {
    content: "",
    tool_calls: [],
    invalid_tool_calls: [],
    additional_kwargs: {
      function_call: { name: "calculator", arguments: '{"input":"2 + 3"}' },
      tool_calls: undefined
    },
    response_metadata: {}
  },
  lc_namespace: [ "langchain_core", "messages" ],
  content: "",
  name: undefined,
  additional_kwargs: {
    function_call: { name: "calculator", arguments: '{"input":"2 + 3"}' },
    tool_calls: undefined
  },
  response_metadata: {
    tokenUsage: { completionTokens: 17, promptTokens: 77, totalTokens: 94 },
    finish_reason: "function_call"
  },
  tool_calls: [],
  invalid_tool_calls: []
}

message.additional_kwargs["function_call"];

{ name: "calculator", arguments: '{"input":"2 + 3"}' }

Or we can use the update OpenAI API that uses tools and tool_choice instead of functions and function_call by using ChatOpenAI.bind({ tools }):

const modelWithTools = model.bind({
  tools,
});
await modelWithTools.invoke([
  new HumanMessage("use a calculator to add 2 and 3"),
]);

AIMessage {
  lc_serializable: true,
  lc_kwargs: {
    content: "",
    tool_calls: [
      {
        name: "calculator",
        args: { input: "2 + 3" },
        id: "call_R2WXRRIbgh2qn1XBBu8P91DH"
      }
    ],
    invalid_tool_calls: [],
    additional_kwargs: {
      function_call: undefined,
      tool_calls: [
        {
          id: "call_R2WXRRIbgh2qn1XBBu8P91DH",
          type: "function",
          function: [Object]
        }
      ]
    },
    response_metadata: {}
  },
  lc_namespace: [ "langchain_core", "messages" ],
  content: "",
  name: undefined,
  additional_kwargs: {
    function_call: undefined,
    tool_calls: [
      {
        id: "call_R2WXRRIbgh2qn1XBBu8P91DH",
        type: "function",
        function: { name: "calculator", arguments: '{"input":"2 + 3"}' }
      }
    ]
  },
  response_metadata: {
    tokenUsage: { completionTokens: 17, promptTokens: 77, totalTokens: 94 },
    finish_reason: "tool_calls"
  },
  tool_calls: [
    {
      name: "calculator",
      args: { input: "2 + 3" },
      id: "call_R2WXRRIbgh2qn1XBBu8P91DH"
    }
  ],
  invalid_tool_calls: []
}