Version 3.13.5

October the 24th

• [Devops]: We have provisionned more ressources to the LLM based NLU as more customers are using it
• [Conversations]: We have fixed an issue where some shared conversations took too much time to load
• [Knowledge Base]: Ability to upload up to 100 files in a row.

Version 3.13.4

October the 23th

• [API]: Added the chunks retrieval endpoint which looks for relevant text chunks in all the knowledge bases of your assistant. more on this here. This endpoint is really interesting if you simply want to check what documents are matching with your asssistant's knowledge base. This already the API used in the Chunk Search Engine

• [UI]: Show AI providers logos as well as the base model as often as possible
• [Wording]: the LLM section has been renamed "AI Deployments" to avoid ambiguity with the "Base model" names which are set by the AI providers.

• [UI]: IDs are shown as often as possible so that you can know the ID of a knowledge base, an assistant, a model, ect for easier use with the API
• [Datasets]: In the QnA section, as more and more customers are using LLM based bots that relies on the intent description, we have added a column named "Description" to the data table for more visibility

Version 3.13.3

October the 21th

• [API]: Added an endpoint to delete knowledge base documents from the platform disk. More on this here.
• [Builder]: Fixed an issue on phone call buttons, you can now freely use any phone number, including non numeric codes and DTMF
• [Builder]: Fixed some issues with URL buttons in carousels
• [Collaboration]: We added the support emails to the contact list so you can share your bot with us in case of support need. Contact us if you have any pb!

• [Knowledge Base]: Open the file browser directly once the "Add file" button is clicked
• [Evaluations]: New module to create and run evaluations on the bots with the "LLM as a Djudge" technique

• [Evaluations]: Added an export button to get the results in an Excel friendly format
• [Evaluations]: Improved prompt to get the best possible Comparaison Eval. For each test, the returned score is a similiraty rate. In our global score calculation, and because the eval is made for a regression test purpose, only bot responses with a 100% similarity eval will be accepted as succesful test.
• [Evaluations]: Added access to the LLM as a Djudge comment as a tooltip of the similarity score

Version 3.13.2

October the 16th

• [Builder]: Fix an issue where pressing Ctrl Shift could make the arrows disappear
• [Bot Response Translation]: Cache management to avoid retranslating the same bot answer again and again.

Version 3.13.1

October the 11th

• [Tools on Agents]: This allows your agents to execute code!
• [NLU]: New kind of intents => Logical Intents
• [Chunks]: Show line returns for better readability

Version 3.13.0

October the 10th

• [Bot sharing]: Ability to share a bot easily with your team mates thanks to user suggestions

• [UI]: Added a copy button to all platform instructions for easier duplication.

• [Webchat]: Ability to restrict the display of the webchat to specific domain names only (e.g., https://my-intranet.my-domain.com). This is very convenient when dealing with sensitive data or to avoid hackers to reuse it in a fake website and do fishing attacks/

Upcoming features for this month (work in progress)

to be released

• [Models]: Addition of OpenAI o1 et o1-mini models
• [Callbots]: Connect your agent to a phone number using the new real time audio api
• ...