ElevenLabs configuration

ElevenLabs agent settings include speech keyterms, interruption ignore terms and languages, turn timing, filler messages and background voice filtering. The interruptions switch changes the interruption client event while preserving other events. Voice speed accepts values from 0.7 to 1.2.

Configure built-in tools under the agent prompt: end call, transfer to number, transfer to agent, voicemail detection, keypad tones, language detection and skip turn. Tool configurations use the provider JSON editor and are validated before deployment. Imports preserve language-preset and workflow prompts.

Evaluation criteria prompts and descriptions of existing named data-collection fields accept registry prompts. Post-call webhook settings require an ElevenLabs workspace webhook ID. Setup webhook request headers may reference an organization secret using ${SECRET_NAME} or Bearer ${SECRET_NAME}; deployment substitutes a workspace secret locator. Privacy, retention and guardrails are available in the provider configuration.

Prompt-only deployments preserve live speech and tool settings. Verification compares authored configuration against the latest agent version and excludes server-added defaults.

Optional agent name, tags and platform settings accept the provider's nullable values. Native MCP server settings include secret-backed URLs, request headers and request metadata. Header and metadata keys retain their original spelling. Deprecated provider fields remain available for imported configurations.

Legacy MCP configurations support headers and timeoutSeconds both at the top level and inside server. Native top-level requestHeaders and responseTimeoutSecs take precedence over those aliases. An explicit transport takes precedence over legacy metadata.protocol; sse selects SSE and shttp selects streamable HTTP.