v244 Easy RAG with new `LLM Say` block π§
"I may be synthetic, but I'm not stupid." - Bishop, Aliens π€
SentiOne Automate version 244 was released on 7.11.2024. This latest release brings a brand new "LLM Say" block, that significantly simplifies the creation of RAG-based bots
LLM Say
You can now use OpenAI ChatGPT completion API directly from the block, without the need to configure custom REST Integration. This feature and Document Semantic Search built-in Knowledge Base allow you to create state-of-the-art bots based on the Retrieval Augmented Generation (RAG) paradigm.
The most basic RAG flow
You can download sample bot project here with data that can be indexed in Knowledge Base. Remember to set your OpenAI credentials in Organization settings!
π IMPROVEMENTS
- [Flows] Export your flow canvas to .png file!
- [OpenAI] New knowledge.semanticSearch function that used credentials stored in company settings
- [EntraID] Use secure Azure EntraID to log in your Automate deployment
- [Voice] Errors from Voice Gateway are logged and visible in Conversation module
- [Channels] Display reason of channel's invalidation
- [LiveChat] Support for command BlockingOperation
- [LiveChat] Setting custom variables from Bot with SetLiveChatCustomVariables command
- [Knowledge Base] Renaming of "AI Items" to Documents and Semantic Search
- [NLU] Added logging of processing time per each NLU service in application nlu-pipeline
π FIXES
- [Flows] Versions were not displayed correctly in history
- [Optimization] Front-end kept calling conversation details endpoint after user has left Conversations module
- [Flows] Copying block with tags between projects didn't work correctly
- [WebChat] In certain conditions webchat would display the Agent's reply from another conversation
- [Conversations] When SentiOne Listen was selected as source, conversation wouldn't close when filters were changed and visible conversation would not match search criteria
Please follow changelog for exiting new features! π