Release notes – 3.4.0

Release date: 21/08/2025 (early morning)

Reworked admin panel

We redesigned the admin panel for knowledge bases and generative responses. You can now choose different LLM models for specific tasks (answers, contextualizer, and answer checking). Each LLM integration you add can be selected via a dropdown for these functionalities, which must still be enabled in your bot settings to take effect. The datasources panel has also been reworked for more clarity and ease of use, especially when working with multiple datasources.

• How it works: Go to your bot’s admin panel, open the LLM hub, add your desired LLM integrations, and select them in the dropdowns for each functionality. Make sure to activate the desired features in the bot settings.

LLM contextualiser (V1.1)

When using a knowledge base or search with a generative chatbot, follow-up questions (e.g., “Can you expand on that?”) often return poor results due to a lack of context. With the new contextualizer setting, the bot rewrites such queries using the conversation history to create a better search query for the knowledge base.

• How it works: Enable the contextualizer in your bot settings and select the LLM model you want to use for it in the LLM hub.

LLM answer check

Generative scenarios previously only entered the “bot not understood” flow if an error occurred. LLMs often produce an answer regardless of accuracy, which may not always be desirable. The answer check feature adds an extra LLM call to verify whether the generated answer is correct and relevant, helping to catch hallucinations and mark the correct status.

• How it works: Enable the answer check option in your bot settings and select your preferred LLM model for the check in the LLM hub.

Improvements

• Zendesk datasource update: Added more generic fields for improved flexibility (no setup changes needed).

• Genesys integration: Improved stability and enhanced some functionalities.

• LLM deletion safeguard: You now get an error if you try to delete an LLM integration that is still in use.

Bug fixes

• Bot export/import training bug: Fixed an issue where exporting and importing a bot could disable the training button.

• Entity timestamp issue: Corrected timestamp mismatches that caused training sentences to load incorrectly into memory.

• Copy update: Updated text in the retraining button for clarity.

What this means for you

If you are using generative chatbots, you can now improve search accuracy with the contextualizer, ensure response reliability with answer check, and have more control over which LLMs power each part of your bot’s process. These changes don’t affect non-generative bots or customers not using these features. Contact us if you need guidance on enabling or configuring them.