Changelog
AppService status
Back to All

v261 New Intent Models & Enhanced API 🔍

1 day ago by Michał Brzezicki

"Affirmative, Dave. I read you." - HAL 9000, 2001: A Space Odyssey 🤖
SentiOne Automate version 261 was released on 07.05.2026. This release introduces new embedding-based intent recognition models, enhanced ASR provider support, and expanded Public API capabilities for conversation management.

New Intent Recognition Models

Five new embedding-based models are now available for intent classification, offering a range of performance and accuracy options:

  • Paraphrase ML Mini LM L12 V2 - Fastest multilingual model (response time < 15ms)
  • Multilingual E5 Small - Balanced multilingual model (response time < 20ms)
  • Multilingual E5 Base - Enhanced accuracy (response time < 50ms)
  • BGE M3 - Most accurate multilingual model (response time < 120ms)
  • Auto - Recommended option with automatic model selection (longer training time)

These models provide better multilingual support and faster inference times compared to traditional approaches.

Enhanced Voice Channel Support

ElevenLabs ASR: New support for ElevenLabs speech recognition with configurable language codes (ISO 639-1/639-3 format) across 30+ languages.

Deepgram ASR: Full integration with Deepgram's speech recognition, including Nova-3 and Nova-2 models, with configurable endpointing and numerals settings.

Google ASR Fix: Resolved configuration issues in Additional ASRs where Google provider was missing required configuration tags.

Expanded Public API

New conversation management endpoints provide programmatic access to conversation data:

  • Conversation Search (POST /api/public/v1/conversations/search) - Search conversations with flexible filters including project IDs, session IDs, time ranges, and cursor-based pagination
  • Conversation Messages (GET /api/public/v1/conversations/{internalSessionId}/messages) - Retrieve complete message history with NLU results, commands, tags, and metrics
  • Recording Download (GET /api/public/v1/conversations/{internalSessionId}/recording) - Download audio recordings for voice conversations

Full API documentation is available at automate.help.sentione.com/reference.

Flow Execution Improvements

Automaton Execution Logic: Flow execution now prevents the same automaton from executing multiple times in a single turn, improving predictability and reducing unexpected behavior when returning from contexts or redirected flows.

Stack Overflow Protection: Enhanced handling of flow stack overflow scenarios with graceful error reporting, preventing bot crashes when flows contain excessive redirect loops.

Additional Improvements

  • Test Suite Debugging: Failed command assertions now display actual commands and tags received, making test failures easier to diagnose
  • WebChat Multi-line Input: Message input field now supports multi-line text entry for longer messages
  • PATCH Method Support: Integration blocks now support HTTP PATCH method for API calls

💊 IMPROVEMENTS

  • [NLU] Five new embedding-based intent recognition models with multilingual support
  • [Voice] ElevenLabs ASR support with 30+ language codes
  • [Voice] Deepgram ASR integration with Nova models and configurable settings
  • [API] Public API conversation search endpoint with comprehensive filters
  • [API] Public API conversation messages retrieval endpoint
  • [API] Public API recording download endpoint for voice conversations
  • [Integration] PATCH HTTP method support in Integration blocks
  • [Flow] Automaton execution limited to once per turn for more predictable behavior
  • [Flow] Enhanced stack overflow protection with graceful error reporting
  • [WebChat] Multi-line message input field for longer user messages
  • [Test Suites] Actual commands and tags displayed when assertion fails

🐛 FIXES

  • [Voice] Google ASR now works correctly in Additional ASRs with proper configuration tags

  • [UI] Flow errors no longer disappear when switching between contexts

  • [UI] Condition block expand icon fixed - no more text overlap or incorrect hover states

  • [UI] Intent label updates now refresh immediately without requiring page reload

  • [UI] Integration block advanced authentication settings now preserved when updating

  • [UI] Organization Secrets validation messages display correctly without character splitting

  • [UI] Sandbox UI form compatibility improved

  • [Infrastructure] Dual Redis write capability for Knowledge Base documents during migration

Follow the changelog for more exciting new features! 👋