Documentation
FAQ
Enum

Frequency Asked Questions

Enum

General

What is Enum

EnumTech(R) is a Canada-based company that has a SaaS-based product called Enum https://enumhq.com (opens in a new tab). We help individuals and companies use AI technologies for their business, work, entertainment, and beyond.

When we launched

We launched in May 2023.

How to contact you to get support or submit feedback/suggestions/feature requests?

For simple questions, you can try our own chatbot on enumhq.com (opens in a new tab), and for anything else please write us at info@enumhq.com.

Payments and subscriptions

If I use my own OpenAI key on a "Key" plan, why should I pay you?

You pay a small fee that partially covers the costs of running our service. You probably don't know but under the hood, it's a very complicated and pretty expensive system containing many elements such as databases, hosting our web application, and connecting to other services (for example, in some cases, we use a paid crawling service to make the process of crawling websites smoother).

AI and ChatGPT

Where do I enter an OpenAI key?

You enter it in your account's settings; it's located in the left menu. It's only available for plans starting with "Key".

image

Is it safe to enter my OpenAI key and then save it in your system?

We store all such information (like the OpenAI key and your password encrypted). Of course, nowadays, there is always some risk of data breach but currently, it would be pretty hard to crack all these keys - we use the SHA-256 algorithm which is almost impossible to crack.

If I use my own OpenAI key, is there any difference in the responses?

When you use your own OpenAI key, you also can choose another model, say ChatGPT-4. This model shows better results (but it's much more expensive). If you stay with the GPT-3.5 model, the result will be the same.

Where do I select an OpenAI model to be used?

It can be selected in 2 places:

  1. In your account settings:

image

  1. In your project's settings:

image

Why do we have these settings in two different places? You may want to set a model for most of your projects, like a default one, and then set a different model for a specific project.

Which AI do you use?

We generate answers via OpenAI ChatGPT (opens in a new tab).

Which models do you use?

To create embeddings, we use the text-embedding-ada-002 model. For generating responses, we use the gpt-3.5-turbo, gpt-4, and gpt-4-turbo models. Please note that two latter models are available for users who bought the "Key" plan. Also, lower models like Curie and DaVinci are available.

The chatbot's responses are too short (or too verbose), how to fix it?

You may create a custom prompt where you describe the style of responding. For example, if the responses are too short you may write "Respond in a polite, verbose manner explaining every single detail" or something like that. This setting is available on the project's settings page, on the AI Settings tab:

image

Here you also can check the Replace the default prompt switch (if it's not turned on, your custom prompt will be added to our default prompt).

What is the default prompt and should I replace it or not?

The default prompt is not just a prompt, it's a combination of prompts and code to help achieve the best results while using the ChatGPT. It has such features as language detection and others.

Can the chatbot keep a casual conversation?

Yes, it can. You may ask it about common things, greet, or even invite for a date.

Crawling and indexing

When I crawl my website your system can't find any pages or too few, why?

There could be several reasons for that:

  1. Your website is protected too hard. We already cope with CloudFlare protection but it may be even harder. Check with your provider if they protect your website from crawling by bots.
  2. Your website is dynamic that is the content is loaded by JavaScript dynamically which currently, we are not able to read.
  3. Check if your website should (or should not) use the "www" prefix.

When I started crawling/indexing, everything went okay but then I got an error

If you see an error message while indexing, it may mean that something went wrong. Possible reasons:

  • you lost the Internet connection
  • there was an error on the server
  • the server was down
  • your pages are protected

Although there are several reasons, none of them is fatal for your indexing process. During the crawling and indexing process, we update your data source on a regular base, so no data is lost. During the indexing, we also uncheck every URL that was indexed so if the process suddenly stops you can continue with other pages. To continue, just refresh the page in your browser, and click the Keep crawling or Start indexing - whatever you started, the corresponding process will start from the point where it stopped.

In the case of website protection, unfortunately, we can't help with it. We use the special crawler if we see any signs of protection but it doesn't always help. If your pages are hosted with Zendesk, please read their comments on this issue: https://support.zendesk.com/hc/en-us/community/posts/4412354085658-SEMRUSH-crawl-returns-403-errors-on-Zendesk-links (opens in a new tab)

Using the chatbot

I've created and indexed my chatbot but when I asked it, it gave irrelevant answers or said "I didn't know" even though I provided all the information in the documents.

This problem may have several causes. Let's review them:

  • Wrong or confusing prompt

ChatGPT doesn't "think" as we do - it "understands" simple constructions and may be confused with lists. You can get more pieces of advice from our article (opens in a new tab)

  • Fine-tuning parameters don't fit your documents

We have a couple of parameters that can change the results dramatically. Both can be found in the Fine tuning section of the project's settings:

image

The chunk size is the size of a piece of text that your primary documents are split into. When we index your documents, we always split them into chunks. The default value is 256 (number of tokens, read more here https://platform.openai.com/tokenizer (opens in a new tab)) but more or less value may give better results. It depends on the density of your text (some texts are pretty sparse which means they have more spaces and new lines therefore the bigger chunks would work better.

💡

The only way to apply this setting is to re-index your data source.

The number of chunks to be sent as context can change the results too. Fortunately, there is no need to re-index your data sources to apply this setting. Try to change the number and see what is the result. You can test querying your data in the Interactions (opens in a new tab) as it's the fastest way to make sure your changes work.

My chatbot only answers "Let me send your enquiry to our colleague who may know the answer"

This answer means you reached the limits of usage. Please check your limits on the Limits/Usage page (the link to it can be found on the left bottom menu).

Enum plugin for Crisp

General questions

How to install the Crisp chatbot on my website?

Please follow the Crisp guide (opens in a new tab).

I can't see the Enum plugin on Crisp's marketplace anymore, what does it mean?

Sometimes (pretty rarely, though) we started working on big changes that can't be deployed immediately. Crisp may require us to hide the plugin from the market if it's not production-ready. If it happens it means we will be there again as soon as we prepare the plugin to be ready for public.

Does the chatbot understand other than English, languages?

Yes, it does. It automatically detects the language and answers in the corresponding language. It also translates the buttons. Currently, we can translate in 107 languages.

How do I connect the Crisp chatbot and Enum?

Please follow our guide on installing the Crisp plugin (opens in a new tab).

My chatbot isn't working!

There are several reasons why it may not work (by working we mean the chatbot is answering your users' questions).

  1. Check if the Enum plugin for Crisp was installed and integrated successfully. If you see this, it means you actually didn't install the plugin:

    image

  2. Check if your chatbot is enabled.

image

  1. Check if you have more than other projects and accounts with the same chatbots. We do not prohibit creating multiple accounts with us (if it doesn't break our terms) but it happens with some of our users that they forgot about already existing accounts with the same bot. When a new message is coming our inner application looks for any project by the website id and there is no guarantee it will be a "right" project. So, if it's your case, just sign in using another browser, then find the project and remove the chatbot you don't want to be confused with.

  2. Check your OpenAI account (for those who are on the "Key" plan). If you are out of your credits, no more calls to ChatGPT can be made until you pay (and you should pay in advance, they don't have "pay-as-you-go" option). You can look at the project's logs to see if there are error messages like this one:

image

  1. Check if your sessions are in the list of paused sessions (the corresponding tab of your chatbot). The chatbot can be paused if you set up the corresponding flags and they worked, for instance, when an agent joins the conversation, the chatbot can be paused.

  2. Check the schedule for your chatbot - it may be exact the timeslot when it's paused.

  3. Check the Integration with Crisp chatbot workflow setting for your chatbot. If you've selected the second option, the chatbot will always wait for a signal from the Crisp chatbot to start, else it won't start. The first option is the default one and doesn't prevent the chatbot from answering.

image

  1. Check if you exceeded your Enum limits, you can find it on the Limits/Usage page accessing from the left menu:

image

  1. There could be outages of OpenAI. Check this page https://status.openai.com/history (opens in a new tab) to see if it's the case.

  2. Check if you set the chatbot to run only in the test environment. This setting can be found on the Live test tab of your chatbot:

image

How to prevent the chatbot from interrupting the conversation with an agent?

There is a setting you can use to pause the chatbot as soon as an agent joins the conversation. It can be found in the chatbot's settings, on the Custom Behaviour & Automation tab, in the Chatbot in session tab:

image

The icon (avatar) is not showing for my chatbot after I changed it

There could be a couple of problems with your new icon (avatar):

  1. First of all, currently Crisp supports only image values. So, .PNG, .JPG (or .JPEG), .BMP are valid extensions for your images. But embedded images are not working. Also, we currently take only URLs which means the images should be uploaded to some server.
  2. Crisp has a pretty load timeout time (the waiting time) to render the icon, so the image should be pretty small for fast load. Also, your server should be fast enough. So, if you don't know how to resize the image or have doubts about your server's speed, please send us your picture at info@enumhq.com and we upload it to our server (the GitHub) which is proven to be fast.