New LLM Support and Agent Features

ElevenLabs has expanded its LLM provider support to include Claude Sonnet 4.6 and Gemini 3.1 Flash Lite Preview for agent conversation configuration, giving developers access to latest generation models. Additionally, the platform now exposes agent versioning tracking with enable_versioning and enable_versioning_if_not_enabled query parameters on agent creation and update endpoints.

Advanced Guardrail Controls

Custom guardrails now support configurable execution modes through the new GuardrailExecutionMode enum, with separate input and output schema definitions. This enables developers to specify precisely when guardrail logic runs and which message direction it applies to. A new guardrail_triggered client event has been added to the ClientEvent enum, allowing agent workflows and client-side handlers to respond when guardrails fire during conversations.

Batch Calling and Telephony Improvements

The batch calling API now includes a target_concurrency_limit field, enabling you to control the target number of concurrent calls for a given batch. Telephony outbound calls for both Twilio and SIP trunk now support the new telephony_call_config field with a configurable ringing_timeout_secs (default: 60 seconds). WhatsApp integration gains enable_audio_message_response support, allowing agents to send audio messages alongside text in conversations.

Speech-to-Text and SDK Updates

The speech-to-text API now offers a no_verbatim transcription option, allowing the model to produce cleaned-up output rather than strict verbatim transcripts. JavaScript and Python SDKs have been updated to v2.38.0+ with schema regeneration and bugfixes, including CRLF parsing corrections for the music detailed endpoint. The convai-widget packages (v0.10.1+) address compatibility issues with Wix sites where addEventListener is non-writable.

Mobile and Voices Enhancements

Enterprise customers can now leverage SAML SSO for sign-in across iOS and Android apps. Voice captcha rendering has been fixed for non-Latin scripts (Hebrew, Thai, Armenian, Georgian, Malayalam, Telugu, Gurmukhi), with proper font support and language-aware selection replacing English defaults.

API Changes Summary

Key endpoint updates include:

• Agents API: New versioning controls, test filtering by type (deprecated include_folders)
• Speech-to-Text: Added no_verbatim option; removed scribe_v2_turbo
• Studio: New agent_settings and playback_speed fields
• Audio Native: Made author and title optional in updates
• New schemas: TelephonyCallConfig, GuardrailExecutionMode, StudioAgentSettingsModel, and SMB tool parameter types