CustomGPT

CustomGPT client library for nodejs. You need the customgpt service running to use this library. You can reach out to me privately at ([email protected]) to purchase a licence.

Features

• Train ChatGPT with your data
• Supports Clip vision (Can detect image links in your content)
• Supports text-davinci-003, text-davinci-002, gpt-3.5-turbo and gpt-4 models

Install

npm install @davmixcool/customgpt

Quick Start

Use it in your project:

const { CustomGPT } = require("@davmixcool/customgpt")

const customgpt = new CustomGPT({
    key: "YOUR API KEY",
    host: "http://localhost:3000/"
});

Train CustomGPT with your data

Here is a basic example that trains CustomGPT with your data. It assumes the CustomGPT service is running at localhost:3000.

We create a collection to segment our data

//Supply the collection name to be created
let collection = await customgpt.create_collection("articles");
if (collection.err) {
    console.error(collection.err);
} else {
    console.log(collection.response);
    // {
    // "id": "aaff97d6-e9a5-11ec-91ba-d954177814f8"
    // }
}

Next we train customGPT with our data.

let payload = {
    collection_id: "aaff97d6-e9a5-11ec-91ba-d954177814f8",
    content: `My quest for a good SEO meta tag implementation in Laravel drove me tech mad to write a package that will add standard SEO meta tags to my application with ease. However, I had to go the extra mile to research important meta tags and the role they play when it comes to SEO and how they can be used to improve SEO, So literally I had to do most of the heavy lifting. Lets quickly take a detour to what meta tags are and how they can improve SEO before we unveil the package.`,
    tags: ['SEO'] //An array of tags used to categorise the source
}
let training = await customgpt.start_training(payload);
if (training.err) {
    console.error(training.err);
} else {
    console.log(training.response);
    //A source id is returned used to update and manange training
    //
    // {
    // "id": "c26b16b4-d394-11ed-b5a3-33d8a09a24e3"
    // }
}

We can also update a training with it's source id

let payload = {
    source_id: "c26b16b4-d394-11ed-b5a3-33d8a09a24e3",
    content: `My quest for a good SEO meta tag implementation in Laravel drove me tech mad to write a package that will add standard SEO meta tags to my application with ease. However, I had to go the extra mile to research important meta tags and the role they play when it comes to SEO and how they can be used to improve SEO, So literally I had to do most of the heavy lifting. Lets quickly take a detour to what meta tags are and how they can improve SEO before we unveil the package. Why do meta tags matter? As previously mentioned, meta tags offer more details about your site to search engines and website visitors who encounter your site in the SERP. They can be optimized to highlight the most important elements of your content and make your website stand out in search results. Search engines increasingly value good user experience, and that includes making sure that your site satisfies a user’s query as best as it possibly can. Meta tags help with this by making sure that the information searchers need to know about your site is displayed upfront in a concise and useful fashion.`,
    tags: ['SEO',"Meta"] //An array of tags used to categorise the source
}
let training = await customgpt.update_training(payload);
if (training.err) {
    console.error(training.err);
} else {
    console.log(training.response);
    // {
    // "id": "c26b16b4-d394-11ed-b5a3-33d8a09a24e3"
    // }
}

We can then chat with our trained data

let payload = {
    collection_id: "aaff97d6-e9a5-11ec-91ba-d954177814f8",//The collection to chat
    openai_key: "YOUR OPENAI API KEY",
    model: "gpt-3.5-turbo", //The model to use for the chat response. Defaults to gpt-3.5-turbo. Options: gpt-3, gpt-3.5-turbo, gpt-3.5-turbo-0301 or gpt-4
    message: "What are meta tags?", //The chat message
    source_ids: ["c26b16b4-d394-11ed-b5a3-33d8a09a24e3"], //An optional array of source ids to chat. Will chat the whole collection if not provided
    tags: ["SEO"], //An optional array of tags to filter by
    top: 10, //The number of vector embeddings used to form a context. A lower number between 5 - 10 gives a more specific answer. Defaults to 5.
    max_tokens: 300, // The maximum number of tokens allowed for the chat response. By default, the number of tokens the model can return will be (300 - tokens).
    instruction: "You are an AI assistant, a creative business assistant that completes requests and always formats his responses in HTML. You are my friendly business AI assistant that is very informative & creative and can provide advice or complete creative tasks that I request. You use the information in the knowledge base as context if relevant. When you respond to me, your answer must be formatted in HTML so it is easier to read with paragraph tags, line breaks, heading and bold for titles, and use lists or tables when applicable." //An instruction to tell the AI how to reply
}
let chat = await customgpt.chat(payload);
if (chat.err) {
    console.error(chat.err);
} else {
    console.log(chat.response);
    // {
    //     "reply": "The reply goes here.",
    //     "meta": [
    //         {
    //             "source_id": "c26b16b4-d394-11ed-b5a3-33d8a09a24e3",
    //             "tags": [
    //                 "SEO",
    //                 "Meta"
    //             ]
    //         }
    //     ],
    //     "usage": {
    //         "prompt_tokens": 763,
    //         "completion_tokens": 86,
    //         "total_tokens": 849
    //     }
    // }
}

Delete a Collection

//Supply the collection id to delete
let collection = await customgpt.delete_collection("aaff97d6-e9a5-11ec-91ba-d954177814f8");

if (collection.err) {
    console.error(collection.err);
} else {
    console.log(collection.response);
    // {
    // "message": "Deleted succcessfully"
    // }
}

Delete a Training

//Supply the source id to delete
let training = await customgpt.delete_training("c26b16b4-d394-11ed-b5a3-33d8a09a24e3");

if (training.err) {
    console.error(training.err);
} else {
    console.log(training.response);
    // {
    // "message": "Deleted succcessfully"
    // }
}

Conventions

All methods must be awaited, and return a CustomGPTResponse object - which only has two properties: err and response.

Always check for presence of err. If err is not null, then the response might not be valid.

Methods

With a customgpt object, just await one of the following methods to interact with the CustomGPT service:

create_collection(collection_name)

Creates a new collection with collection_name

delete_collection(collection_id)

Deletes a collection and all its trained data with the collection_id

start_training(payload)

Trains CustomGPT with your data

let payload = {
    collection_id: "",//The collection to create the source in
    content: "", //The content to train CustomGPT on
    tags: [""], //An optional array of tags used to categorise your source source
}

update_training(payload)

Updates training source

let payload = {
    source_id: "",//The source source to update
    content: "", //The content to train CustomGPT on
    tags: [""], //An optional array of tags used to categorise your source source
}

get_training(source_id)

Gets the embeddings of a training with the source_id

delete_training(source_id)

Deletes a training with the source_id

delete_batch_training(payload)

Deletes a batch training using a collection_id and the training source_ids in a payload

let payload = {
    collection_id: "",//The collection
    source_ids: [""], //An array of source ids
}

chat(payload)

Chat your trained data with a payload

let payload = {
    collection_id: "",//The collection to chat
    openai_key: "YOUR OPENAI API KEY",
    model: "gpt-3.5-turbo", //The model to use for the chat response. Defaults to gpt-3.5-turbo. Options: gpt-3, gpt-3.5-turbo, gpt-3.5-turbo-0301 or gpt-4
    message: "", //The chat message
    source_ids: [""], //An optional array of source ids to chat. Will chat the whole collection if not provided
    tags: [""], //An optional array of tags to filter by
    top: 10, //The number of top vector embeddings used to form a context. A lower number between 5 - 10 gives a more specific answer. Defaults to 5.
    max_tokens: 300, // The maximum number of tokens allowed for the chat response. By default, the number of tokens the model can return will be (300 - tokens).
    instruction: "" //An instruction to tell the AI how to reply
}

search(payload)

Search your trained data with a payload

let payload = {
    collection_id: "",//The collection to search
    keyword: "", //The search keywords
    source_ids: [""], //An optional array of source ids to search. Will search the whole collection if not provided
    tags: [""], //An optional array of tags to filter by
    top: 20, //The number of top vector embeddings to return.
}

Maintainers

This package is maintained by David Oti!

License

This package is licensed under the Apache license.