v249 New LLM Integration block 🔌

“This change will be good for all of us." - Cassandra, Cassandra (2025) 🤖
SentiOne Automate version 249 was released on 24.04.2025. This new release introduces a new block type called LLM Integration.


Use LLM the way you want!

The LLM Integration block allows you to send a custom query to an LLM of your choosing (ChatGPT, Gemini, etc.) and process the results within the bot. Unlike LLM Say, it does not stream the response to the user but stores it in memory. With this new feature, you can:

  • Check user input for specific characteristics
  • Use an LLM as an alternative to the built-in NLU
  • Create custom guardrails to ensure the conversation aligns with bot policy

LLM Integration works similarly to standard Integration blocks, but it saves configuration time and reduces the risk of faulty HTTP calls.


💊 IMPROVEMENTS

  • [Voice] New option in Microsoft Speech-to-Text engine: “Silence timeout after user’s utterance”
  • [Voice] Raw mode is now available, allowing the use of non-standard parameters with ASR and TTS engines
  • [AI Credentials] Vendor icons now appear in lists
  • [AI Credentials] Improved validation to check whether provided credentials are functional
  • [Security] Default session timeout updated to match OWASP ASVS requirements. Default values are described here

🐛 FIXES

  • [Admin] Fixed unnecessary highlight on the left-side menu visible on macOS
  • [Analytics] Fixed freezing of relative filters that caused incorrect display for "last 7 days"
  • [OpenAI on Azure] Fixed validation error for valid endpoint URLs and added support for non-default deployment names
  • [Multi-language] AI-generated response variants are now translated into other languages defined in the flow
  • [Analytics] Fixed granularity issues when the time range spanned more than one hour
  • [Channels] Fixed issue where users without channel-edit permissions encountered errors in the Flow module
  • [Conversations] Fixed issue where metadata values that were arrays only showed the first element in the preview
  • [LLM Say] Fixed issue where default AI credential and model disappeared after saving an LLM Say block
  • [Chat Tester] Fixed issue where entity summaries showed "undefined" for detected sys.Time entities

Follow the changelog for more exciting new features! 👋