The docs are still a work in progress and not all pages are written yet! But don't worry, if the page you're looking for is not documented yet, you can check out the auto-generated docs. There is a lot to cover, so documentation will be added over time!
What is GPT Turbo?
GPT Turbo is a JavaScript library (with TypeScript-first support) for interacting with OpenAI's Chat Completion API. It focuses on covering 100% of the API surface, while providing a simple, intuitive, and type-safe interface.
Features
🤖 Supports all Chat Completion models, including GPT-4. (full list here)
💬 Supports both single, streamed and function completions, just like ChatGPT.
⚙ Tune chat completion parameters, such as temperature, top-p, and frequency penalty.
🌐 Compatible in both Node.js and the browser.
📜 Keeps track of the conversation history for you, making conversation continuity a breeze.
💰 Estimate the cost and size of conversations before sending them to the API. (through the gpt-turbo-plugin-stats plugin)
💾 Easily persist conversations with serialization and deserialization methods.
🔌 Includes a plugin system for extending the library's functionality.
💻 Built entirely with TypeScript.
⚔️ Battle-tested in multiple environments. (See implementations)
Implementations
Here's a list of all the home-made implementations of GPT Turbo:
- Web: A web app, very similar to ChatGPT, for handling chats in the browser.
- Discord: A Discord bot for chatting in Discord servers, similar to Discord's own Clyde AI bot.
- CLI: A command-line interface to chat straight from your terminal.
- Nest: A NestJS backend, for interacting with the library via a REST API.
Plugins
Here's a list of all the home-made plugins for GPT Turbo:
- gpt-turbo-plugin-stats: A plugin for estimating the cost and size of conversations before sending them to the API.
Differences from OpenAI's official library
While OpenAI provides an official JavaScript library for interacting with the API, it focuses on providing a basic interface for all of their APIs. This is fine when you're working with multiple of them, but not ideal when you just want to use the Chat Completion API. GPT Turbo is designed to be a dedicated library for the Chat Completion API, and thus provides an interface centered around it as well as features like conversation history management.
One of the pain points of working with the API directly or OpenAI's official JavaScript library was keeping track of the message history.
Since you need to send the entire history of messages for each prompt, you need an efficient way to store and update the history.
GPT Turbo solves this by providing a Conversation class that handles all of this for you.
While conversation history management was the main motivation for creating this library, it has grown to do much more than that.
GPT Turbo also supports proxy configuration, message streaming, callable functions and plugins.
Another issue with OpenAI's library is that it is designed as a server-side library, and thus does not support the browser. GPT Turbo is completely isomorphic, and works in both Node.js and the browser.
API Reference
These docs are designed to be clean and concise, and thus do not include the entire API reference. For a complete and detailed reference on the API, check out the auto-generated docs. The website a bit less polished than this one, since it's auto-generated with TSdoc, but it contains the most up-to-date reference, only for the latest version of GPT Turbo.
Recipes
If you just want to jump into some of the concepts of GPT Turbo by example, check out the recipes section. Many of these were once written directly in the library's README and served as the only source of documentation for their features. So you should be able to follow along without reading through the entire docs!
Acknowledgements
Special mention to:
- Mantine, the UI library used to build this docs website. The docs structure is also heavily inspired by theirs.
- Cuputo on Flaticon for the logo.
next
Installation