Version 3.6.0

🗓 Release dates

• Smartly.AI: 8th
• Virtual Agent: tbd

Hello and Happy New Year! 🎉

As we step into a new year, we're thrilled to kick off with a special release: a Generative AI (GenAI) focused version of our platform. This update is packed with advancements aimed at elevating your experience and enhancing the capabilities of our platform. We're excited for you to explore these new features and improvements as we continue to evolve and grow together in the coming year.

[Generative AI]: Support of New LLM Providers

We're excited to announce support for two leading Large Language Model (LLM) providers, enhancing our chatbot platform's capabilities:

Mistral AI

Our platform now integrates with Mistral AI's models, including the Mistral 7B and Mixtral 8x7B.

• Mistral 7B: This model is equipped with 7 billion parameters, making it exceptionally proficient in generative text tasks. Its design caters to a broad spectrum of applications, delivering robust performance in text generation, making it an ideal choice for diverse conversational scenarios.
• Mixtral 8x7B: As a more advanced option, the Mixtral 8x7B model features 12 billion active parameters. It stands out for its proficiency in handling multiple languages and accommodating larger context windows, ensuring a more seamless and engaging conversational experience across various languages.

Note: Make sure to request your API token from their website: mistral.ai

Microsoft Azure OpenAI

For businesses already leveraging Microsoft Azure, we now provide integration with Azure Open AI. This collaboration allows for seamless use of Azure's AI capabilities within our platform, offering a more secure and private option for enterprises concerned with data privacy and compliance.

• Open AI: You can use your own OpenAI token.

[Generative AI]: Source view

You can understand every bot answer by clicking on the "View Sources" button
The source view is accessible via the "Conversations" module

[Generative AI]: Log view

By clicking on the view logs button, you will get access to a detailed JSON with more insights on how the answer generation occured

The source view is accessible via the "Conversations" module

[Generative AI]: Enhanced Knowledge Base Management

We've restructured the Generative module for more efficient management of knowledge bases, distinguishing them from assistants. This change introduces two separate sections:

• Assistants: This section is dedicated to configuring general settings for your chatbot assistants.
• Knowledge Bases: Here, you can create and manage groups of data sources. These knowledge bases can then be integrated into your assistants as needed.

As shown below, you can organize your content in knowledge bases than you can then attach to your assistant

Streamlined Knowledge Base Creation and Integration

Create as many Knowledge Bases as you need

Add those Knowledge Bases to your assistants

Advantages of the New Structure:

• Focused Data Processing: Update and process subsets of data sources independently within your knowledge bases. This approach eliminates the need for full reindexing every time an update is required, saving time and resources.
• Reusable Knowledge Across Assistants: Index your data once and use it across multiple assistants. This feature allows for more efficient data management and reduces redundancy.
• Customized Settings for Each Knowledge Base: Tailor the settings of each knowledge base to meet specific requirements, such as different scraping or chunking strategies, based on the nature of the documents.

[Generative AI]: Enhanced Generative AI Integration Across Multiple Channels

We are excited to announce that our advanced Generative AI capabilities are now integrated across multiple communication channels. This expansion builds on the existing chatbot functionalities, bringing the sophisticated conversational experiences of our Generative AI to a wider range of platforms.

Delivery Dynamics across Channels:

• Smartly Webchat: In our webchat, the AI-driven responses are uniquely displayed in a progressive, word-by-word manner for an engaging user experience.
• Other Channels: For other platforms like RingCentral, Messenger, and WhatsApp, the delivery of AI-generated messages may differ. Messages can be received either fully upon generation or in segmented chunks, depending on the channel’s specific capabilities and user interaction model.

Version 3.6.1

🗓 Release dates

• Smartly.AI: 19th
• Virtual Agent: 25th

Changes

• Improved: Show the total number of documents coming from all the knowledge bases attached to an assistant
• Improved: Better UI for adding sources in knowledge bases
• Improved: Better management of CSV files
• New: Ability to upload html files in the knowledge bases

Version 3.6.2

🗓 Release dates

• Smartly.AI: 24th
• Virtual Agent: 25th

Gen AI

• Assistant (Removed): Role of the assistant, can be set with the custom instructions
• Documentation (Updated): API endpoints where updated following recent changes in data format. Check it out here
• Knowledge bases (New): Ability to select two different web scrappers: Cheerio more suitable for static pages (the default as it is working with most websites), Puppeteer more suitable for web pages with dynamic content.
• Knowledge bases (Improved): the Cheerio selector allows you to define wich html element you want to scrap specifically
• Knowledge bases (New): Powerpoint document support. Note that this will get the text from the slides and will create one text chunk per slide.
• Knowledge bases (Improved): Ability to visualize the chunks for each document
• Assistant (Improved): When chatting with the assistant, the document search is now faster thanks to a parallelization of the search requests.

NLU (SNP 1.06)

• Improved: Fix all seed values to get repeatable results
• Improved: Better performance for XG Boost classifier