Version 3.14.2

Released on January 17th

Migration completed: All knowledge base vectors have been moved from Pinecone to Elasticsearch, enhancing scalability, retrieval performance, and overall system reliability.

Version 3.14.1

Released on January 15th

Released on January 10th

Introduced visual badges to highlight key information at each processing step.
These badges now display the number of outputs generated, offering a quick way to assess response complexity and relevance.

Users can now choose their own embeddings deployment, offering more flexibility in how knowledge base data is vectorized and processed. If you decided to use your own embeddings, remember to use the same embeddings in your knowledge and your AI Assistant.

Released on January 9th

Added a new report that provides a breakdown of interactions across different processing methods:
- Classical NLU (rule-based intent detection)
- LLM-based NLU (leveraging large language models for understanding)
- RAG-based requests (retrieval-augmented generation queries)
This report helps users analyze the balance between different AI approaches and optimize performance accordingly.
To access this data visualization, go to Analytics > Understanding in the platform.

Introduced a new setting that allows users to keep specific files private, ensuring that they remain visible only to them and not accessible to bot users—even if a direct file link is shared.

Dark mode is now available for the bot platform, providing a more comfortable visual experience, especially in low-light environments.