You’re not really dealing with a language problem , it’s a structure problem. Cloning 8 English assistants into 8 Vietnamese ones will just double complexity and make things harder to manage long term. What you actually need is smarter routing, not more assistants. I’d keep your current setup, add a language detection layer upfront, and route into the right flow without duplicating everything. Much cleaner and scalable. I recently did this for Soltan and Francis , both had similar issues with multilingual + messy assistant setups, and we fixed it without bloating the system. Quick question: Are you detecting language before transcription, and are your assistants role-based or flow-based?