@truto/langchain-toolset NPM

Truto's SuperAI Toolset for Langchain.js

A powerful toolset for integrating Truto's Proxy APIs with Langchain.js applications. This package provides all the Proxy APIs configured to be used as a tool on Truto as tools to Langchain.

Installation

npm install @truto/langchain-toolset
# or
yarn add @truto/langchain-toolset

Prerequisites

Node.js 22.14.0 or higher
Yarn 1.22.19 or higher
Truto API credentials
Bun (for running examples)
OpenAI API key (only needed for running examples)

Environment Setup

Create a .env file in your project root with the following variables:

TRUTO_API_TOKEN=your_truto_api_token

#OPTIONAL
OPENAI_API_KEY=your_openai_api_key # if you are using OpenAI models
TRUTO_API_BASE_URL=your_truto_api_base_url

Usage

Here's a complete example showing how to use the toolset with Langchain.js, including handling multiple tool executions:

import { ChatOpenAI } from '@langchain/openai';
import { HumanMessage, SystemMessage } from '@langchain/core/messages';
import { isEmpty, values } from 'lodash-es';
import { ToolCall } from '@langchain/core/dist/messages/tool';
import { getTools } from '@truto/langchain-toolset';

// Initialize the LLM
const llm = new ChatOpenAI({
  model: 'o3-mini',
});

const INTEGRATED_ACCOUNT_ID = 'your_account_id';

async function main() {
  // Get all available tools for a customer connection
  const tools = await getTools(INTEGRATED_ACCOUNT_ID, {
    truto: {
      baseUrl: process.env.TRUTO_API_BASE_URL,
      token: process.env.TRUTO_API_TOKEN as string,
    },
  });

  // Bind tools to the LLM
  const llmWithTools = llm.bindTools(values(tools));

  // Create your messages
  const messages = [
    new SystemMessage(
      'You are an assistant who uses the available tools to give the user an answer. Make sure you respect the arguments required for a tool call, use them to filter down the results wherever necessary. All the tools return a JSON string response, so parse the output correctly and use them in the arguments.'
    ),
    new HumanMessage('Your question here'),
  ];

  let toolCalls: ToolCall[] = [];
  do {
    // Execute any pending tool calls
    for (const toolCall of toolCalls) {
      const toolResponse = await tools[toolCall.name].invoke(toolCall);
      console.log('\n===================\n');
      console.log(
        'Called tool',
        toolCall.name,
        toolCall.args,
        toolResponse.content
      );
      messages.push(toolResponse);
    }

    // Get the next AI response
    const aiMessage = await llmWithTools.invoke(messages);
    messages.push(aiMessage);

    // Check if there are more tool calls to execute
    if (isEmpty(aiMessage.tool_calls)) {
      console.log('\n\n===================\n\n');
      console.log(aiMessage.content);
      break;
    }
    toolCalls = aiMessage.tool_calls || [];
  } while (!isEmpty(toolCalls));
}

main();

This example demonstrates:

How to initialize the tools with Truto credentials
How to bind tools to a Langchain.js LLM
How to handle multiple tool calls in sequence
How to process tool responses and continue the conversation
How to properly handle the conversation flow until all tool calls are complete

Running Examples

The repository includes example code in the examples directory. To run the examples:

First, install Bun if you haven't already:
- Visit bun.sh to install Bun for your platform
- Or use the following command:
```
curl -fsSL https://bun.sh/install | bash
```
Install dependencies: