ElevenLabs

Music v2 added streaming with metadata. POST /v1/music/detailed/stream SSE-streams composed audio alongside a full composition plan, song metadata, and optional word timestamps via with_timestamps. Both v1 and v2 plan shapes remain accepted.

Agents gained per-agent sentiment scoring and auto-translate. Platform settings added optional sentiment_analysis for post-call scoring per agent and auto_translate_transcript_to_app_language to render transcripts in the viewer's language. Backchannel detection marks filtered utterances as ignored_as_backchannel. Auxiliary audio presence is now required on conversation responses.

Tool interruption mode replaced the binary toggle. disable_interruptions is deprecated across agent tool and MCP server schemas. Use interruption_mode (allow, disable_during_tool, disable_during_tool_and_turn) instead, with per-tool overrides. Also deprecated: the Simulate conversation endpoints — use the Create test and Run tests routes instead.

Nested agent transfers and MCP environment scoping landed. Workflows add push, pop, and replace operations for nested agent transfers, with enable_nesting and return_when_nested controls. MCP server and tool routes added optional environment scoping.

SDKs added an onPing callback for latency visibility. The @elevenlabs/client and @elevenlabs/react packages now expose an onPing callback that surfaces ping events including ping_ms. The SDK still replies to pings automatically. Patch bumps also landed for the convai widget core/embed and React Native SDK.