• Upload Your Knowledge Base
• Supported Knowledge Sources
• URLs
• Sitemaps
• Files
• URL Scraping
• Inspect Your Knowledge Base
• Token Limits
• Retraining Websites
• Deleting Data Sources
• Test Your Knowledge
• Setting Up Your AI Assistant
• Quick Start
• Failure Handovers

Rasayel’s AI Assistant optimized for customer support is available to use as a Chatbot node

By providing your knowledge base to the AI Assistant, may that be your public documentation, help center articles, or any internal SOP documents, the AI Assistant is able to take control of the conversation and reply to your customers on your behalf.

Naturally, like any customer support agent, the AI Assistant will ask clarifying questions if the customer’s inquiry is not clear enough.

Whenever the AI Agent does not have enough knowledge to answer the question, it’ll hand over the conversation based on your setup.

Upload Your Knowledge Base

1. Go to “Automations”
2. Click on “Artificial Intelligence”
3. Add your first data source by clicking on “Add Data Source”

You can ingest different types of knowledge data sources into Rasayel.

Supported Knowledge Sources

URLs

You can add multiple URLs to Rasayel. The URLs must be publicly accessible. Follow these steps:

1. Click on “Add”
2. Fill in the URL to your data source
3. Add as many URLs as you need
4. Click on “Create”

Sitemaps

If your website has an XML sitemap, you can provide the sitemap URL to Rasayel, and Rasayel will automatically ingest all URLs present in the sitemap

1. Click on “Sitemap”
2. Fill in the Sitemap URL
3. Click on “Create”

Files

Rasayel also supports ingesting files as data sources. We support a variety of file types:

• Markdown: md
• Word Documents: doc, docx
• Excel sheets: csv, xlsx
• Text: txt
• PDF: pdf

To add your files:

1. Click on “Files”
2. Click on “Click or drag files to upload”
3. Select the files you want to upload
4. Allow some time for the files to finish uploading
5. Click on “Create”

URL Scraping

You can provide the entry point to your knowledge base/documentation to Rasayel and we will scrape all the sub-websites for you.

URL Scraping will recursively scrape websites from the provided URL up to a depth of 3.

To scrape a website:

1. Click on “URL Scraping”
2. Fill in the URL Scraping entry point
3. Click on “Create”

Inspect Your Knowledge Base

Once you add your knowledge source, it will be listed on the page in a “Training” state

Training knowledge sources can take some time, based on the size of the data. It usually takes up to a few minutes.

If you are using a Sitemap or URL Scraping, you will be able to inspect which websites Rasayel has extracted out of your knowledge base

Each data source will be trained separately, and you will be able to inspect the state of each data.

Once the knowledge source processes completely, it will enter a "Trained" state, at which, you will be able to know how many LLM Tokens we have consumed to train it.

Token Limits

Each data source that you add consumes LLM tokens to train, therefore, it can become expensive for Rasayel to maintain your data sources indefinitely.

For this reason, Rasayel imposes a limit of 10 million tokens/month for each workspace. The amount of tokens consumed by a data source depends on how big the source is.

You can find the current limit consumption on the bottom left of your data sources screen.

Retraining Websites

Naturally, your help center websites may change with time, at which case, you will be able to “Retrain” your data sources.

Simply, click on “Retrain” next to the updated data source. If you Retrain a Sitemap or a URL Scraping data source, it will also retrain all sub-URLs under it.

Deleting Data Sources

If a data source is no longer relevant to your AI Assistant, you can delete it by clicking “Delete”

Test Your Knowledge

After adding your data sources, a good next step is to test the AI Assistant to see how it works with your data. This will help you identify missing knowledge from your knowledge base, allowing you to improve your knowledge incrementally.

You can start testing your AI Assistant Knowledge by clicking on “Test my AI”, under which, a test screen will pop up, prompting you to chat with your AI Assistant.

If you wish to start a brand new session with the AI Assistant, click “Clear” in the top right corner

Setting Up Your AI Assistant

The AI Assistant can be set up as part of any Chatbot flow, (read more about Chatbots ).

Within the right sidebar of the Chatbot Builder, you will find the Support AI node

The Support AI node can be linked after any other node, including the trigger.

By default, the Support AI node contains two out-links:

• Query resolved: when the AI Assistant determines that the contact’s question has been answered.
• Customer frustrated or AI can’t answer: when the AI Assistant does not have enough knowledge to answer the conversation.

Quick Start

Having the AI Assistant as a Chatbot node, allows you to create all kinds of automations that fit your workflow.

But if you are unsure and want to get started quickly, here is how a very basic AI Assistant Chatbot setup can be:

In this setup, the AI Assistant will take over the conversation once a message is received.

The AI Assistant will continue answering the contact until the query is resolved, or one of the handovers is fulfilled.

If the query has been resolved, it will Solve the conversation, therefore, eliminating all noise from your inbox.

If it cannot carry on with the conversation, it will hand over the conversation to your team by opening it.

Failure Handovers

For queries you prefer not to be handled by the AI Assistant, you can add those queries as ‘failure handovers’.

Failure handovers are scenarios that the AI will not handle if they occur. For example, you may not want the AI Assistant to handle any Sales or Billing related conversations, and instead, hand them to the corresponding team.

You can add failure handovers on the node itself, handovers are defined by their ‘name’ and ‘description’.

An example handover can be:

• Name: Sales or Billing
• Description: When the contact asks any questions about sales or has a problem with billing

In this example, we use the Sales or Billing handover to assign the conversation to a specific agent whenever it is sales-related.