Skip to content

Latest commit

 

History

History
172 lines (123 loc) · 6.41 KB

README.md

File metadata and controls

172 lines (123 loc) · 6.41 KB

ChatGPT-Lite: OpenAI REST API standalone wrapper for NodeJS or Web

NPM version

ChatGPT-Lite is a minimal standalone JS wrapper for the OpenAI REST API.

It provides dependency-free access to common OpenAI REST API calls, from Node or Web.

The intention is to cover only the most frequent use cases. No fluff, no filler; so tiny that you can use it inside a Bookmarklet.

This is not trying to be a complete replacement for OpenAI's official library, which is located here: https://github.com/openai/openai-node

Installation

Node via npm:

npm install chatgpt-lite
import OpenAI from "chatgpt-lite"; // ESM.mjs
// const OpenAI = require("chatgpt-lite"); // CJS.js

const openai = OpenAI({ apiKey: "api_key" });

Web via CDN:

<script type="module">
  import OpenAI from "https://unpkg.com/chatgpt-lite/index.mjs";
  // import OpenAI from "https://cdn.skypack.dev/chatgpt-lite";

  const openai = OpenAI({ apiKey: "api_key" });
</script>

or

<script src="https://unpkg.com/chatgpt-lite/index.js"></script>
<script>
  const openai = OpenAI({ apiKey: "api_key" });
</script>

Usage

You can browse the ChatGPT-Lite code examples.

If you wish to expand the core functionality (to add function calls, process streaming, etc.), update/add code to match the OpenAI API signatures -- refer to their official API Reference and Documentation.

The code below shows how to get started, using the core .create() async function similar to the official chat completions API.

import OpenAI from "chatgpt-lite"; // ESM.mjs
// const OpenAI = require("chatgpt-lite"); // CJS.js

async function main() {

  // similar to official API, but no need for `new OpenAI({ ... });`
  const openai = OpenAI({
    // for minimal size it does NOT currently default to setting apiKey = process.env["OPENAI_API_KEY"]
    apiKey: "api_key_can_not_be_omitted"
    // apiKey: process.env["OPENAI_API_KEY"] || await fetchApiKey()
  });
  console.log("openai:", openai);
  // openai: {
  //   apiKey: 'sk-QiyAbcDe1FGhIjk23l4mnOpQ5RstuV67w89x012y2z34Ab7C',
  //   create: [AsyncFunction: create]
  // }

  // same signature/result as `await openai.chat.completions.create({ ... });`
  const completion = await openai.create({
    messages: [{ role: "user", content: "Say this is a test" }],
    model: "gpt-3.5-turbo",
  });
  console.log("completion:", completion);
  // completion: {
  //   id: 'chatcmpl-8xAbCdEFghI4JklmNopQrs7TuvwxYz',
  //   object: 'chat.completion',
  //   created: 1709167986,
  //   model: 'gpt-3.5-turbo-0125',
  //   choices: [
  //     {
  //       index: 0,
  //       message: [Object],
  //       logprobs: null,
  //       finish_reason: 'stop'
  //     }
  //   ],
  //   usage: { prompt_tokens: 12, completion_tokens: 5, total_tokens: 17 },
  //   system_fingerprint: 'fp_86123a45b6'
  // }

  const {usage} = completion;
  console.log("usage:", usage);
  // usage: { prompt_tokens: 12, completion_tokens: 5, total_tokens: 17 }

  const message = completion.error || completion.choices[0].message;
  console.log("message:", message);
  // message: { role: 'assistant', content: 'This is a test.' }

};

main();

Handling errors

Errors might occur when ChatGPT-Lite is unable to connect to the OpenAI API, or if the API returns a non-success status code (i.e., 4xx or 5xx response).

The returned Promise resolves to the {object} returned by fetch's Response.json(). Success means this object will contain a property named choices. Failure means it will contain a property named error and it will be missing the property named choices.

const completion = await openai.create({ model: "gpt-1.2-turboff", messages: [ { role: "user" } ] });
console.log("completion:", completion);
// completion: {
//   error: {
//     type: "e.g. invalid_request_error",
//     code: "e.g. model_not_found",
//     param: "e.g. messages.[0].content",
//     message: "e.g. Invalid value for 'content': expected a string, got null."
//   }
// }
console.log("completion.choices:", completion.choices);
// completion.choices: <undefined>

Streaming Responses

Currently this does not provide support for streaming responses. A stream option might be added in the future.

Automated function calls

Currently this does not provide support for function calls / running tools. This option might be added in the future.

File Uploads

Currently this does not provide support for file uploads. This option might be added in the future.

Retries

Currently this does not provide support for retrying after certain errors. A maxRetries option might be added in the future.

Timeouts

Currently requests should automatically timeout after the number of minutes that the OpenAI API defines as the default (likely 10 minutes, or perhaps 20). A timeout option might be added in the future.

Requests which time out will not be retried by default. This might be added in the future.

Auto-pagination

Currently this does not provide support for any kind of limit option to produce paginated output. This might be added in the future.

Advanced Usage

Accessing raw Response data (e.g., headers)

The "raw" Response returned by fetch() is not returned by any functions. A raw option might be added in the future.

Customizing the fetch client

By default, ChatGPT-Lite uses the built-in fetch() async function provided by Node(v18+, released 2022-04-19) or most modern Web browsers. A fetch option might be added in the future, to allow for providing a custom fetch function, which for example could be used to inspect or alter the Request or Response before/after each request.

Currently this does not look for a DEBUG=true environment variable, but in the future this might be added to log all requests and responses automatically (for debugging purposes).

Configuring an HTTP(S) Agent (e.g., for proxies)

Currently this does not provide support for any kind of baseURL or httpAgent options, for example to use the API behind a proxy. This might be added in the future.

Requirements

In theory, this should work inside any JavaScript runtime environment that provides a built-in fetch() async function. In theory.