Release of April 20th, 2026
✨ New Features
External User Data
Institutions can now connect external data sources to enrich conversation context. Supported sources: CSV upload, Microsoft Entra, and LTI injector parameters. Fields (e.g. student program, campus, enrollment status) can be mapped, aliased, and surfaced in the Inbox. Flow conditions and custom message variables will come in a future release.
📄 “How do I add external user data to an assistant?”
MathML for Accessible Math Rendering
Math expressions now render using the MathML standard, enabling screen readers (VoiceOver, NVDA, JAWS) to read them properly instead of outputting raw markup. You may also notice some early improvements to how math is displayed visually in chat, with a more comprehensive update to follow in a future release.
Assistant-to-Assistant Handover
You can now hand off a conversation from one assistant to another directly within a flow, allowing you to route users to a more specialized assistant based on their needs.
Brightspace LOR PDF Ingestion
PDFs stored in the Brightspace Learning Object Repository and embedded in courses are now picked up during knowledge ingestion.
🛠 Improvements
Flow Email Action — Expanded Metadata
Flow-triggered emails now support a much richer set of template variables: conversation summary, history, user location, launch URL, IP address, and other system metadata.
Style Tab — Header Color Picker & Layout Redesign
Admins can now set a custom background color for the chat header. Text and buttons adapt automatically for readability. The Style tab layout has also been cleaned up with better labeling and spacing.
Voice Mode — Multilingual Support
Voice mode now supports multilingual dictation and audio playback, removing the previous English-only constraint.
AI Feedback Improvements
Several updates this release: automated feedback now has its own independent "Submit as draft" setting; automated entries show "Automated" as the teacher name in the dashboard and exports; teacher thumbs up/down ratings are now included in data exports; and Feedback V2 assistants show an animated thinking steps view during post-generation.