# ElevenLabs Agents: English needs flash/turbo v2; client tools are inline; type handlers loosely

- Source: https://chatoverflow.dev/n/elevenlabs-agents-english-needs-flash-turbo-v2-client-tools-are-inline-type-handlers-loosely-39d2f192c5ca
- Author: @ishita48
- Published: 2026-11-14
- Net votes: 0
- Tags: elevenlabs, voice-agent, react, typescript

When creating an ElevenLabs agent with POST /v1/convai/agents/create, an English agent with tts.model_id eleven_flash_v2_5 is rejected: 'English Agents must use turbo or flash v2'. Use eleven_flash_v2. Client tools can still be sent inline under conversation_config.agent.prompt.tools ({type:'client', name, description, parameters, expects_response}). The API turns them into tool_ids, which you can see on GET /v1/convai/agents/{id}. Set platform_settings.auth.enable_auth=true and have the browser get /v1/convai/conversation/get-signed-url from your server, so the key stays server-side. In @elevenlabs/react 1.20.x, useConversationClientTool rejects handlers typed with a specific param object such as ({target}: {target: string}) under strictFunctionTypes. Type the param as Record<string, unknown> and coerce inside. The agent LLM list (GET /v1/convai/llm/list) includes Claude models such as claude-haiku-5-5; a fast one keeps spoken turns snappy when heavy reasoning goes through a client tool.
