SquirroGPT Release Highlights#

This page details release highlights for SquirroGPT, a retrieval-augmented generation (RAG) AI application that allows you to conversationally interact with your own organizational or website data.

For detailed Squirro platform release notes, see Squirro Release Notes.

November 03, 2023#

This release includes numerous new features and performance improvements, including the following:

  • Digital assistant white labeling improvements

  • Chat histories now saved

  • Under the hood IR improvements

  • New user preferences options

  • New data source overview feature

  • Strict mode grounding by default

  • Prompt guideline documentation

  • An improved web crawler

  • Updated welcome dialog

Digital Assistant White Labeling Improvements#

You now have a wide range of new and improved options for customizing the SquirroGPT digital assistant interface to match your brand.

Clicking Embed Digital Assistant will launch your embed options. After selecting from the available options, you’ll be presented with white labeling customization options, as shown in the example Embed as a Floating Action Button option below:

SquirroGPT Embedding White Label Options

Under the Hood IR Improvements#

SquirroGPT developers and machine learning engineers are constantly working to improve the underlying information retrieval (IR) system. This release includes numerous improvements to the IR system, which will result in better retrieval and grounding. As a new technology, there’s still room for improvement, and the Squirro team is excited to continue to improve the IR system in future releases as well as associated features like highlighting.

Chat Histories Now Saved#

SquirroGPT now saves individual user chat histories. This allows you to view your chat history with SquirroGPT and, using the User Preferences detailed in the following section, download your chat history as a JSON file.

New User Preferences Options#

By clicking the three-dot menu next to the support icon then User Preferences, you now can perform the following actions:

SquirroGPT

  • Export Chat History: Download your chat history as a JSON file by clicking Export Data.

  • Clear All Chat Data: Permanently delete chat data by clicking Clear All Chats.

  • Interface Language: Select the language of the chat interface by selecting from the drop-down menu.

Note

The Interface Language setting only controls the language of the SquirroGPT user interface. It does not control the language of SquirroGPT responses.

New Data Source Overview Feature#

You can now easily view your connected data sources, as well as their status, by clicking the server icon as indicated in the following screenshot:

SquirroGPT Data Sources icon

Click Configure Data Sources to view your connected data sources in the Squirro Setup space and manage them in detail.

Reference: To learn more about loading and managing data sources within the Squirro platform, see How to Load Data Using the UI.

Strict Mode Grounding by Default#

Grounding refers to the process of connecting a user’s query to a specific document in the data source. SquirroGPT now uses strict mode grounding by default, which means that SquirroGPT adheres as strongly as possible to the retrieved sources, limiting answers generated from the LLM’s internal knowledge.

SquirroGPT Strict Grounding Example

Prompt Guideline Documentation#

You can now view prompt guideline tips and documentation by clicking the info icon. This will launch the Prompt Tips window, which includes the View Detailed Guidelines link to the full documentation.

SquirroGPT Prompt Tips

An Improved Web Crawler#

For SquirroGPT Web users, the web crawler has been improved to offer better flexibility in selecting the data you want to crawl.

The Crawler data loading connector now includes the following features:

  • Add Include Path: Specify part of a URL to only crawl pages that contain the inputted path within the URL.

  • Add Exclude Path: Specify part of a URL to exclude pages that contain the inputted path within the URL.

  • Add Xpath to Exclude Content: Specify xpath (e.g. //div[@class=`footer`]) to exclude content from the crawled page. Leave blank to crawl the entire site.

Updated Web Crawler interface

Updated Welcome Dialog#

The SquirroGPT welcome dialog has been updated to remove references to automatic plan upgrades. Free trial plans will now be canceled with notice rather than automatically be upgraded.