Projects

A project is a place where you add data sources of the same topics. For example, you may have a separate project for your project, a website, or a group of websites.

Add a new project

To add a new project, click the My projects menu item on the left menu bar, then click the All projects link and the Add a project button.

In the showing popup enter the name and description (optional), then click the OK button:

You will be redirected to the project's page with an empty list of data sources.

Add a data source

As soon as a project is created you may add the data source. A data source is a piece of information to be indexed and searched by our engine. Read more on the data source documentation page (opens in a new tab).

Click the "Add a new source" button and then select the corresponding type of data sources:

Project's settings

After a project is created you may edit some of its properties. For a given project, click the Settings link:

General tab

Here you can change the project's name and its description. The description is used when the project is shown on the Project list:

AI Settings

The second tab has the AI tuning settings.

Let's review all the settings

ChatGPT Engine and OpenAI Model

Here you can select what to use for the generating responses: the OpenAI assistant, a new technology that allows you to upload your own data right in the OpenAI dashboard (opens in a new tab), or a regular OpenAI chat model. If you use the second option, you can then select a model (the default model is ChatGPT-3.5-Turbo). The GPT-4 and GPT-4 Turbo are available.

Custom prompt

This section has 2 options:

• the text of your custom prompt
• the switch to replace the default prompt (that one we use by default) with your prompt.

Also, to help you with creating a good, working prompt, we wrote this article: Art of Creating Great Prompts (opens in a new tab).

Interactions

Here, you control your interaction with AI in the Interactions window.

Specify, if you want the sources to be shown in the interaction window, how many and which types of.

Here is an example of querying with sources:

Custom messages

Currently, there are 4 custom messages you can specify for your chatbot:

1. What the bot says when cannot find an answer
2. What the bot says when a user asks to talk to a human
3. What the bot says when you reach the Enum limits
4. What the bot says if there is a server or OpenAI error

The third message is useful for a case when you are out of limit with Enum, but don't want your users to be aware of this fact. The default value is "Let me send your enquiry to our colleague who may know the answer."

Fine-tuning

These settings are used for the fine-tuning data to be passed to ChatGPT as context.

• chunk size is the amount of text in of tokens (opens in a new tab) to be stored and to be obtained from the index database to form context data (the default number is 300 tokens which usually works great but in some cases, another amount, say, 1024 tokens works better)
• the amount of chunks to send as context - the number of pieces that will be sent as context. The more the number the better but sometimes it can be more than the OpenAI threshold which can cause an error.

Stick to context

If you want the chatbot to stick to your context and ignore questions that may be outside of it, use this parameter:

If our system didn't find any relevant information in your data, and this switch is on, the chatbot will answer with a text you set up in the custom messages (opens in a new tab) or with our standard message if this is not defined. Each piece of data obtained from your data has its own relevancy score (the less the more relevant). Usually, a score more than 0.25 is considered irrelevant. But you may control the threshold and include the data with a score higher than 0.25 - find the best one by experimenting.

Using metadata

You now can add metadata in the prompt, currently, the language of the question is available. If you want the chatbot to answer always in the language of question (that is detected automatically), use the following syntax:

... Answer in the {language} language.

Just add the "language" word wrapped with curly brackets, and then, during the preprocessing of your prompt, it will be converted into a normal string (the exact value varies according to the question's language):

... Answer in the French language.

Deleting project

If you scroll down to the end of the settings page you will see the "danger zone":

Click the Delete project button if you'd like to delete a project but be aware that all the associated data sources as well as interactions will be deleted too.

Events logging

We log errors occurring for a project. For example, they include errors related to the OpenAI API such as insufficient credits or the wrong OpenAI key. You can find the log entries in the Settings, logs, stats menu for your project:

Statistics

Right now, we can collect the stats on using the buttons you attach to your messages; we also count the number of the conversations. How to configure the buttons please read the chatbot documentation (opens in a new tab). To allow collecting these stats the corresponding flag should be on (opens in a new tab).

You can find the stats on the menu Settings, logs, stats.

It's pretty clear what all settings mean:

You can choose to see stats on one specific chatbot or all of them. You may see the stats daily, weekly, or monthly. The last switch allows you to choose between counting all the clicks or only the last one for sessions. It means that if a user clicks buttons several times during the session, only the last one will be counted.

You also can have data on each of your chatbots side-by-side (if you have installed many of them) by selecting the Group by chatbot option.