This release focuses on three things that matter in daily work: a clearer Knowledge Base experience, smoother widget behavior (especially on mobile), and a cleaner set of AI actions and settings across the platform.
Knowledge Base: Preview is Now Streaming
The Knowledge Base preview now supports streaming responses.
Instead of waiting for the full answer to appear at once, you can see the response build in real time, which makes testing faster and more transparent.
Also included:
- Improved dialogs for adding content to the Knowledge Base, with a cleaner and easier flow.
- No more duplicated unanswered questions, fixing a bug where the same unanswered query could appear multiple times.
- New controls inside content details to update:
- HTML tags
- URL sync frequency
- content tags
AI Models: More Options, Cleaner Behavior in Settings
We updated the AI model experience across the dashboard:
- New LLM models added and AI models list updated.
- You can now select between multiple LLMs directly from the AI settings dialog.
- Fixed a bug where OpenAI models appeared in the dropdown even when they were not available through environment configuration.
Ask Knowledge Base specific fixes:
- Fixed an issue where Ask KB showed the selected template default value and max-token incorrectly.
- Fixed a bug where editing Ask KB text could cause the selected Knowledge Base value to be lost.
MCP Tools: Selection and Preview Are More Reliable
We improved MCP tool management to reduce friction when building tool-connected AI Agents:
- Fixed issues in selecting and refreshing MCP server tools.
- Updated CSS for the preview of selected tools, making it clearer and more consistent.
Widget Behavior: Better Loading, Smarter Auth, Stronger Mobile UX
This release includes several changes that make the widget feel more predictable and user friendly.
Loading and start behavior
- Default
autoStartis now false, so the widget does not auto-open unless configured to do so. - Added an opening loading spinner when the widget is launched.
- The widget can now load without authentication and still display the speech bubble, so users see the entry point immediately.
Authentication rules are now more consistent
Authentication is now triggered when it is truly required by configuration:
- Start with authentication if
hasCalloutInWidgetConfigis true. - Force authentication if visibility rules are set to OPEN on desktop or mobile.
Mobile behavior
- Mobile always opens fullscreen and ignores legacy stored size.
- The resize-widget button is hidden on mobile to avoid UI clutter.
Conversation UX
- Updated the user typing indicator behavior.
- Added a clear “thinking” indicator while the bot is generating a response.
Design Studio and Flow Builder: Cleaner Actions, Better Logs
We made several improvements to reduce confusion and improve editing speed.
Flow log improvements
- In the Flow log action, error types are now sorted and shown with icons and colors in the select box.
Action list cleanup
- Removed the ChatGPT task action from the list of actions.
- If it is already used in an existing flow, it is now marked with a deprecated tag.
Editing and layout fixes
- Fixed drag behavior so actions can be dragged more reliably, including outside the intent block.
- Improved CSS for actions that use AI Settings.
- Hide the rules button when no rules are set.
- Removed “manage kb” from the webhook detail panel.
Conversation Quality: Better UI for Attachments and History Filters
- Attachment buttons in messages now respect container width and wrap long labels instead of being clipped.
- Added a filter in History to show only rated conversations, making it easier to review feedback signals.
Additional fixes that improve reliability
- Fixed GPT Task preview being broken.
- Fixed condition blocks not saving when changing boolean operators.
- Fixed OpenAI models appearing in selects even when missing from environment variables.
Thanks for reading and for following our progress.
If you want a quick walkthrough of the new features and how to apply them, schedule a demo
