AI Agent

AI Agent simplifies the implementation and use of generative AI with LangChain, you can add components such as vectorized search services (check options in "link"), conversation history (check options in "link"), custom databases (check options in "link") and API contracts (OpenAPI).

Installation

Use the package manager npm to install AI Agent.

npm install ai-agent

Simple use

LLM + Prompt Engineering

const agent = new Agent({
  name: '<name>',
  systemMesssage: '<a message that will specialize your agent>',
  llmConfig: {
    type: '<cloud-provider-llm-service>', // Check availability at <link>
    model: '<llm-model>',
    instance: '<instance-name>', // Optional
    apiKey: '<key-your-llm-service>', // Optional
  },
  chatConfig: {
    temperature: 0,
  },
});

// If stream enabled, receiver on token
agent.on('onToken', async (token) => {
  console.warn('token:', token);
});

agent.on('onMessage', async (message) => {
  console.warn('MESSAGE:', message);
});

await agent.call({
  question: 'What is the best way to get started with Azure?',
  chatThreadID: '<chat-id>',
  stream: true,
});

Using with Chat History

When you use LLM + Chat history all message exchange is persisted and sent to LLM.

  const agent = new Agent({
    name: '<name>',
    systemMesssage: '<a message that will specialize your agent>',
    chatConfig: {
      temperature: 0,
    },
    llmConfig: {
      type: '<cloud-provider-llm-service>', // Check availability at <link>
      model: '<llm-model>',
      instance: '<instance-name>', // Optional
      apiKey: '<key-your-llm-service>', // Optional
    },
    dbHistoryConfig: {
      type: '<type-database>', // Check availability at <link>
      host: '<host-database>', // Optional
      port: "<port-database>", // Optional
      sessionTTL: '<ttl-database>' // Optional. Time the conversation will be saved in the database
      limit: '<limit-messages>' // Optional. Limit set for maximum messages included in conversation prompt
    },
  });

  // If stream enabled, receiver on token
  agent.on('onToken', async (token) => {
    console.warn('token:', token);
  });

  agent.on('onMessage', async (message) => {
    console.warn('MESSAGE:', message);
  });

  await agent.call({
    question: 'What is the best way to get started with Azure?',
    chatThreadID: '<chat-id>',
    stream: true,
  });

Using with Vector stores

When using LLM + Vector stores the Agent finds the documents relevant to the requested input.
The documents found are used for the context of the Agent.

Example of the concept of vectorized search

  const agent = new Agent({
    name: '<name>',
    systemMesssage: '<a message that will specialize your agent>',
    chatConfig: {
      temperature: 0,
    },
    llmConfig: {
      type: '<cloud-provider-llm-service>', // Check availability at <link>
      model: '<llm-model>',
      instance: '<instance-name>', // Optional
      apiKey: '<key-your-llm-service>', // Optional
    },
    vectorStoreConfig: {
      type: '<type-vector-service>', // Check availability at <link>
      apiKey: '<your-api-key>', // Optional
      indexes: ['<index-name>'], // Your indexes name. Optional
      vectorFieldName: '<vector-base-field>', // Optional
      name: '<vector-service-name>', // Optional
      apiVersion: "<api-version>", // Optional
      model: '<llm-model>' // Optional
      customFilters: '<custom-filter>' // Optional. Example: 'field-vector-store=(userSessionId)' check at <link>
    },
  });

  // If stream enabled, receiver on token
  agent.on('onToken', async (token) => {
    console.warn('token:', token);
  });

  agent.on('onMessage', async (message) => {
    console.warn('MESSAGE:', message);
  });

  await agent.call({
    question: 'What is the best way to get started with Azure?',
    chatThreadID: '<chat-id>',
    stream: true,
  });

Using with Database custom

SQL + LLM for prompt construction is a concept that involves using both Structured Query Language (SQL) and LLMs to create queries or prompts for data retrieval or interaction with databases. This approach leverages the power of SQL for database-specific commands and the capabilities of LLMs to generate natural language prompts, making it easier for users to interact with databases and retrieve information in a more user-friendly and intuitive manner.

Example of the concept of SQL + LLM

const agent = new Agent({
  name: '<name>',
  systemMesssage: '<a message that will specialize your agent>',
  chatConfig: {
    temperature: 0,
  },
  llmConfig: {
    type: '<cloud-provider-llm-service>', // Check availability at <link>
    model: '<llm-model>',
    instance: '<instance-name>', // Optional
    apiKey: '<key-your-llm-service>', // Optional
  },
  dataSourceConfig: {
    type: '<type-database>', // Check availability at <link>
    username: '<username-database>', // Require
    password: '<username-pass>', // Require
    host: '<host-database>', // Require
    name: '<connection-name>', // Require
    includesTables: ['<table-name>'], // Optional
    ssl: '<ssl-mode>', // Optional
    maxResult: '<max-result-database>', // Optional. Limit set for maximum data included in conversation prompt.
    customizeSystemMessage: '<custom-chain-prompt>', // Optional. Adds prompt specifications for custom database operations.
  },
});

// If stream enabled, receiver on token
agent.on('onToken', async (token) => {
  console.warn('token:', token);
});

agent.on('onMessage', async (message) => {
  console.warn('MESSAGE:', message);
});

await agent.call({
  question: 'What is the best way to get started with Azure?',
  chatThreadID: '<chat-id>',
  stream: true,
});

Using with OpenAPI contract

OpenAPI + LLM for prompt construction is a concept that combines OpenAPI, a standard for documenting and describing RESTful APIs, with large language models (LLMs). This fusion allows for the automated generation of prompts or queries for interacting with APIs. By using LLMs to understand the OpenAPI specifications and generate natural language prompts, it simplifies and streamlines the process of interfacing with APIs, making it more user-friendly and accessible.

Example of the concept of SQL + OpenAPI

const agent = new Agent({
  name: '<name>',
  systemMesssage: '<a message that will specialize your agent>',
  chatConfig: {
    temperature: 0,
  },
  llmConfig: {
    type: '<cloud-provider-llm-service>', // Check availability at <link>
    model: '<llm-model>',
    instance: '<instance-name>', // Optional
    apiKey: '<key-your-llm-service>', // Optional
  },
  openAPIConfig: {
    xApiKey: '<x-api-key>', // Optional. Using request API
    data: '<data-contract>', // Require. OpenAPI contract
    customizeSystemMessage: '<custom-chain-prompt>', // Optional. Adds prompt specifications for custom openAPI operations.
  },
});

// If stream enabled, receiver on token
agent.on('onToken', async (token) => {
  console.warn('token:', token);
});

agent.on('onMessage', async (message) => {
  console.warn('MESSAGE:', message);
});

await agent.call({
  question: 'What is the best way to get started with Azure?',
  chatThreadID: '<chat-id>',
  stream: true,
});

Contributing

If you've ever wanted to contribute to open source, and a great cause, now is your chance!

See the contributing docs for more information

Contributors ✨

JP. Nobrega 💬 📖 👀 📢

License

Apache-2.0