This release focuses on something teams usually realize too late: running AI agents in production is less about building flows and more about control, visibility, and day-to-day operations.
With this update, you get better ways to manage permissions, understand how your Knowledge Base is performing, and quickly spot what users are asking (and where your AI is failing). On top of that, we cleaned up several rough edges across the widget, flow editor, and conversation handling.
New features
Custom roles for better access control
You can now create custom roles tailored to your team structure.
Instead of forcing everyone into predefined roles, you can define exactly who can access what — whether it’s conversations, settings, or specific features. This is especially useful for larger teams where responsibilities need to be clearly separated.
Read more about the Custom Roles here.
Knowledge Base KPIs at a glance
The Knowledge Bases page now includes KPI cards that give you an immediate snapshot of performance:
- Total contents
- Answered questions
- Unanswered questions
- Answer rate
No more guessing if your AI is doing a decent job. You can actually see it.
Visibility into questions asked to AI agents
You can now view and search the questions users ask your AI agents.
This gives you direct insight into real user intent, helping you understand:
- What people actually ask
- Where your AI performs well
- Where it struggles
Search for unanswered questions
Unanswered questions are now searchable by text.
Instead of scrolling through a messy list, you can quickly find patterns, prioritize gaps, and decide what content to add next to your Knowledge Base.
Ratings directly in the history view
A new rating column has been added to the History table.
User feedback is now visible directly where conversations are listed, making it easier to spot low-quality responses without digging into each interaction.
Quotas reset date visibility
You can now see when your quotas reset, making usage planning a lot less guessy and slightly less annoying.
Improvements
ChatGPT Task action deprecated
The ChatGPT Task action has been removed from the action list.
If you’re already using it in existing flows, it will still appear, but now clearly marked as deprecated, so you know it’s time to migrate.
In its place, we’ve introduced the AI Prompt action, which not only covers the same capabilities as the ChatGPT Task, but goes further:
- Supports more flexible and advanced prompt configurations
- Works with more AI models
- Can connect to MCP tools, allowing your AI agent to execute actions directly based on the prompt
This means your flows are no longer just generating responses; they can actually perform tasks, trigger operations, and integrate with external systems in a more structured way.
Clearer AI settings configuration
Tooltips in the AI Settings dialog have been improved to make configuration less confusing.
Which is good, because nobody enjoys guessing what a parameter does and hoping for the best.
Better flow log readability
Error types in the Flow log action are now:
- Sorted
- Displayed with icons
- Color-coded
So instead of scanning raw chaos, you can actually understand what went wrong.
AI models update
The available AI models have been updated, improving overall performance and keeping things aligned with newer capabilities.
Bug fixes
A mix of fixes that remove small but very real annoyances:
- Teammates with the Agent role can now correctly access unassigned conversations
- Fixed widget loading issues caused by blob handling in
launch.js - Resolved user typing issues when a human agent is involved
- Fixed header menu not working properly on mobile
- Corrected minimum tokens handling when disabling “Get contents sources”
- Improved detection between bot and human messages
- Fixed dragging actions outside of intent blocks
- Resolved CSS issues in actions using AI Settings
Thanks for reading, hope you find our latest improvement helpful.
