GPT Trainer Q&A Document

What is GPT-trainer?

GPT-trainer is a no-code chatbot builder that allows you to train a chatbot using your own data. You can
upload documents or specify a URL to scrape content directly. GPT-trainer has many customization
options and can be embedded directly into your own website. In the near future, GPT-trainer will also
integrate with popular document hosting platforms and collaboration tools like Google Drive and Slack.

How do I use GPT-trainer?

To get started, simply register for an account by clicking “Get started” found on our home page. Once
you have an account, you can create your first chatbot by clicking the button near the center of the
screen.

To upload reference documents for your chatbot, go to the “Manage sources” tab. You can then upload
files directly from your computer, import from an authenticated cloud drive, specify a URL to scrape, or
specify a series of pre-canned responses via text input directly.

After you specify source content, go to “Settings” tab to customize your chatbot. In this tab, you can
manage the general persona of the chatbot, edit the base prompt, specify rate limits, modify the
appearance, and set the website in which you would like to embed the bot.

Once you finish configuring the chatbot, you can test the bot by going to the “Chat” tab.

To embed a chatbot onto your website, select the chatbot you want, then click “Embed on website” near
the top right of the screen.

How much does GPT-trainer cost?

GPT-trainer is free to try. Simply register for an account by clicking “Get started” found on our home
page. If you like the tool, you can upgrade to a paid plan. You can find detailed pricing information by
following this link: https://gpt-trainer.com/#pricing

Where is my data stored?

We are a Software-as-a-Service. This means that our app, along with data you upload to us, resides
online. We use Amazon Web Services (AWS) for hosting. Our servers are located in Oregon, USA.

Does GPT-trainer support other languages?

Yes. Our service currently uses OpenAI's GPT-3.5 and GPT-4 large language models (LLMs) for generative
AI functionality. The models are trained on publicly available data across the internet in over 95
languages, so GPT Trainer also supports over 95 languages.

What should my data look like?

We support most text document formats (.pdf, .docx, .txt, .md, .tex, .html). You can also provide an URL
for automatic scraping of text content (this is not automatically updated on target website refresh) or
input your own text directly.

Does GPT-trainer use GPT-3.5 or GPT-4?

Our back end is designed to be versatile in Large Language Model (LLM) selection. Currently, we have
GPT-3.5-turbo and GPT-4 as options, but we plan on adding more LLMs (e.g., Google Bard, Meta LlaMA,
Anthropic Claude, open-source alternatives, etc.) when they become more mature.

Can I share a chatbot I created?

Input-wise, you are the only administrator to the design, customization, and data of your chatbot.
Output-wise, you can share your chatbot for anyone to use.

Can I give my chatbots instructions?

Yes. You can customize your chatbot to have different personas via our template-guided prompt
engineering.

How can I add my chatbot to my website?

You can embed an iframe or add a chat bubble to the bottom right of your website. To do that, create a
chatbot and click "Embed on website".

How can I reach your team?

You can reach out to us directly via email at [email protected].

How do I delete a chatbot?

You can delete a chatbot by opening the chatbot selection menu on the top right of your screen after
you sign in. You should see a “trash” icon next to each existing chatbot. Clicking on that icon will delete
your chatbot permanently.
Can I delete my account?

You can request a permanent deletion of your account by emailing [email protected]. Please note
that permanent account deletion will immediately terminate any ongoing subscriptions.

I’d like to request a refund!

For billing and support related inquiries, please contact our team via email at [email protected].

Can I sell chatbots I made?

For commercial licensing of our technology or any derivatives therein, please contact our team at
[email protected].

Can I get a free trial of a paid plan?

GPT-trainer comes with a free plan with limited features and message credits. If you would like to expand
the scope of your free plan or request a trial of one of our paid plans, please reach out to hello@gpt-
trainer.com.

What is Petal?

Petal is a sibling product to GPT-trainer that is developed and maintained by the same people. You can
check us out at https://www.petal.org/.

Here are some key points about privacy:

- We are a Software-as-a-Service. This means that our app, along with data you upload to us, resides
online.

- We use Amazon Web Services (AWS) for storage and computation.

- We use OpenAI's GPT-3.5 and GPT-4 language models. This means that text content within your
uploads may, in unmarked chunks, be submitted to OpenAI as context when generating a response. We
will never send your documents in their complete original form to any other organization, including
OpenAI or its partners.

- OpenAI has explicitly stated that it will not use data submitted by customers via its API endpoints to
train or improve models (https://openai.com/policies/api-data-usage-policies).

- Your data is not end-to-end encrypted while stored within our servers. This is because encryption of
your data precludes us from being able to extract any meaningful information from it, thereby disabling
pretty much all our features. This is true for text analytics platforms in general, including OpenAI itself.
Anyone that claims to have end-to-end encryption while offering you similar capabilities is lying.

- Our engineers have access to documents you upload. This is for debugging and user experience
optimization purposes. All our engineers sign legal NDAs so they will not distribute your data in any way.

- We will never sell your data.

- For detailed information regarding our privacy policy or terms of service, please visit https://gpt-
trainer.com/privacy and https://gpt-trainer.com/terms.

What technologies does GPT-trainer use?

GPT-trainer uses a combination of document extraction, chunking, embedding, semantic search, and
large language models (LLMs) from partner service providers to enable its chatbot technology.

Does GPT-trainer fine-tune the large language model (LLM) used to enable generative AI capabilities?

No, GPT-trainer does not fine tune any commercially available large language models (LLMs). It uses
clever document extraction, chunking, embedding, and prompt engineering to anchor chatbot responses
to identified relevant context. Neither we nor our LLM provider partners use your data for LLM training
or fine-tuning purposes.

Can users check chat history?

Bot creators have access to all chat history. Users of the bot do not have access to chat history.

Can the bot collect information from users?

Yes, bot creators can set the bot to collect information such as email, name, phone numbers, etc. from
users of the bot.

Do you offer API access?

No, we do not offer API access at this time. However, our sibling product, Petal, offers white-labeled
solution of the same technology that supports API access. Using the API, you will be able to bring “chat
with your documents” functionality into your own software platform. Please email [email protected] to
consult about licensing options.

Is there a limit on amount of data users can upload?

Depending on your current subscription plan, you will have different limits to the amount of data the bot
can use as context. For details on pricing, please visit https://gpt-trainer.com/#pricing.

Do you offer chatbot analytics?

Bot creators can export entire chat history into a structured Excel document, which can then be used for
analytics purposes by external tools.

How can I make sure my chatbot only uses content from resources that I uploaded?

GPT-trainer’s default prompt sets the bot’s behavior such that it only uses information you provide as
context via uploaded documents, imports from cloud drive, designated URLs, or pre-canned responses.
The bot will say “I don’t know” instead of make things up under this default behavior. You can change
this default behavior by customizing the base prompt.

Does GPT-trainer integrate with Oracle?

Sorry, GPT-trainer does not currently integrate with Oracle. If this is of interest to you and you have a
particular use case in mind, please reach out to us via [email protected] and we can set up an
exploratory call.

Can GPT-trainer handle escalations to live agents?

Sorry, GPT-trainer does not currently automatic escalations to live agents. If this is of interest to you and
you have a particular use case in mind, please reach out to us via [email protected] and we can set
up an exploratory call.

How many credits will I be charged per query?

The message credit cost per query depends on the large language model you selected for your bot to
use. In general, GPT-3.5 uses 1 message credit per query, GPT-3.5-16k uses 8 message credits per query,
and GPT-4 uses 20 message credits per query. This is because more advanced models cost more to
operate.

What if GPT-trainer outputs wrong information?

GPT-trainer’s default setting sets the bot’s behavior such that it only uses information you provide as
context. If it outputs wrong information, you can set up corrections by going to “Manage sources”, then
“Q&A” tab. Here, you can specify any number of pre-canned responses that override the bot’s default
behavior should a query semantically match one or more of your pre-canned questions.
How long are the subscription plans?

GPT-trainer offers both monthly and annual subscription options. You can choose which you prefer when
you upgrade your account.