Release notes - 3.4.2

Release date: 08/10/2025 (early morning)

LLM prompt dynamic values

You can now use curly brackets ( {{ }} ) to call environment variables or context variables in generative scenarios. This makes it easier to dynamically include variables, improving the flexibility of generative flows.

• How it works: In a prompt, you can write for example: The name of our chatbot platform is {{chatbot_name}}. If chatbot_name is defined as a context or environment variable, the value will automatically be inserted. Context variables take priority if there is overlap. If the value is not found, the placeholder remains visible (e.g. {{chatbot_name}}).

Exporting chats

Admins can now export chats directly from the statistics page. You can generate a month-by-month export of all sessions, grouped by session and containing every message for the selected month.

• How it works: Scroll to the bottom of the statistics page and select the month you want to export. A private download link will be generated, available for 48 hours. After that, both the link and export file expire. This feature is only available to organisation and chatbot admins.

Unanswered questions

We have updated the “Train / Unanswered Questions” page to better reflect the state of your chatbots. Alongside standard unanswered questions, you can now also see:

• Generative scenarios flagged as unanswered by the question check feature

• Generative scenarios that ran into technical issues such as LLM guardrails or token limits

• How it works: Go to the Train > Unanswered questions page to review both classic unanswered items and generative-specific issues.

Improvements

• LLM hub design: Navigation within the LLM hub has been streamlined for a smoother experience.

• Error messages: We improved error logging and descriptions, especially for code responses, making them clearer and less ambiguous.

Bug fixes

• Human takeover text: Fixed an issue where text during a human takeover was being split incorrectly.

• Scenario editor drag & drop: Fixed a bug that prevented responses from being reordered via drag and drop.

Coming next

We are working on a major update to the training mechanism, scheduled for release on 29/10/2025.

The retraining process is being completely redesigned to be more efficient and effective, resulting in a smoother experience when configuring chatbots. More details will follow soon.

What this means for you

This release brings helpful improvements to generative flows, unanswered question monitoring, and admin-level chat exports. These changes do not affect your existing chatbot setup.

As always, no action is required on your side. If you have any questions, feel free to reach out to our support team.