hello

[gusty matrix]

AI-generated Answer for baymurat1

Hi! How can I help you with ElevenLabs today? Tell me what you’re trying to do (e.g., generate TTS, clone a voice, use the API), and I’ll point you to the exact docs and examples.

[civic tide]

how llm condition works in workflow?

[gusty matrix]

AI-generated Answer for baymurat1

Short answer: “LLM conditions” in ElevenLabs Workflows are natural‑language rules you attach to transfer tools. At runtime, the agent’s LLM reads those condition strings plus the tool’s description and current conversation context, and if it infers a match it calls the relevant system tool (e.g., transfer_to_agent or transfer_to_number). This is LLM‑driven routing (not a deterministic boolean expression).

Key points

• Where you use them:
  • Agent transfer: each rule has an Agent target and a Condition (NL text). Options include delay_ms, transfer_message, and whether the transferred agent should speak first. The LLM decides when/which agent to transfer to based on those conditions. (elevenlabs.io)
  • Transfer to human: each rule has a Phone Number (E.164) and a Condition (NL text). The LLM chooses when/which number to transfer to. Currently requires Twilio for both the source and target numbers. (elevenlabs.io)
• How the LLM triggers actions: system tools are exposed to the model; when the condition is met, the LLM issues a function call for the tool with required parameters (e.g., agent_number for transfer_to_agent). (elevenlabs.io)

If you meant a different “LLM Condition” node outside of transfer tools, I couldn’t find official documentation for it. Please share a link or clarify the exact UI page/feature. (elevenlabs.io)

Sources:
Agent transfer | ElevenLabs Documentation
Transfer to human | ElevenLabs Documentation
Agent transfer | ElevenLabs Documentation
Transfer to human | ElevenLabs Documentation

[civic tide]

seems it does not have any logic behind, it just comparaes the words. For example, i wrote in llm condition: The user speaks Mandarin or Chinese in their response

however, when the user speaking in chinese does not change the agent