Facing issue:
Agent is not switching language even on consumer’s request.
Details:
- I have added all node prompts, examples, global prompt in English.
- I have selected MultiLingual
- I also have kept guardrails in Global Prompt as follows:
## LANGUAGE & MIRRORING
- strictly detect_language only for: Hindi, Hinglish, Marathi, Gujarati, Tamil, Telugu, Kannada, Odia, Malayalam, Bengali, Urdu, English (India).
- Guest mentions any of the above language, switch to the mentioned language immediately
- after switching-language continue in same language, unless guest requests otherwise.
- Mirror caller language immediately.
- Match formality, pace, and emotional energy.
Emotional & Energy Alignment:
- Excited → warm and enthusiastic tone.
- Calm → composed and clear.
- Hesitant → reassuring and slower.
- Urgent → concise and efficient.
I asked the same question to ASK AI, below is the response I got:
Based on the information sources, language switching in Conversation Flow agents requires specific handling because language switching only happens for the current node—when the agent transitions to another node, it reverts back to English.
Solution for Conversation Flow Agents:
To maintain language consistency across nodes, you need to:
-
Extract the user’s language preference into a dynamic variable (e.g.,
{{user_language}}) after the first node where language selection occurs. -
Add language instructions to every subsequent node. Here’s the recommended prompt structure to add at the top of each node:
Always speak in {{user_language}} throughout this node.
Assume the user only understands {{user_language}}.
Only listen for and recognize speech in {{user_language}}.
Transcribe and interpret the caller's response exclusively in {{user_language}}.
Do not use English unless {{user_language}} is explicitly set to "English".
- Update your global prompt to include:
**Speak only in the user's language ({{user_language}}).**
Alternative Approach:
If maintaining language-specific prompts in every node becomes too complex, you can use agent swap to swap to an agent configured for the specific language. However, this would require maintaining separate agent copies for each language.
Important Note:
When you select the language of the agent in global settings, you’re setting what language the agent will understand (the language of user audio). To make the agent respond in a specific language, you need to write instructions in the prompt such as “respond in Spanish”.
For additional support with your specific configuration, check out the documentation or ask the community for help.
NOT SATISFIED with the AI Response:
- Both the recommended approach, increase TOKEN COUNT Drastically and are not practical.
- I have 40+ NODEs handling various scenarios.
- What is the role of Global Prompt, if we cannot control/manage NODEs centrally?
- Why is 2nd approach recommended, when we have MultiLingual as an option?
error sign (warring or alert symbol)
READERS ALERT 
Do NOT mark any thread as SOLVED unless I consumers confirm. This is so UNETHICAL and MISGUIDING to readers.