Nixflex supports voice agents in 15+ languages. The engine uses a hybrid STT routing strategy and matches the caller’s language automatically.Documentation Index
Fetch the complete documentation index at: https://docs.nixflex.com/llms.txt
Use this file to discover all available pages before exploring further.
How language is decided
The agent’slanguage field is the default — what the agent speaks in if the caller’s language can’t be inferred.
In practice, language is selected as follows:
Caller wins
If the caller speaks a language clearly, the agent matches them — even if
language is set to something else.Country prefix hint
For inbound calls from GCC region numbers (+971, +966, +974, +965, +973, +968), the engine routes to Deepgram for Arabic-first transcription.
Multilingual STT
For all other regions, AssemblyAI’s multilingual model handles 6 major languages (EN, ES, FR, DE, IT, PT) with auto-detection.
Supported languages
Voice (TTS)
15 languages via Inworld TTS Max:| Language | Code | Notes |
|---|---|---|
| English (US/UK/AU) | en | Multiple voice options |
| Arabic | ar | Modern Standard Arabic |
| Spanish | es | |
| French | fr | |
| German | de | |
| Italian | it | |
| Portuguese | pt | |
| Dutch | nl | |
| Polish | pl | |
| Russian | ru | |
| Hindi | hi | |
| Korean | ko | |
| Japanese | ja | |
| Chinese | zh | Mandarin |
| Turkish | tr |
Speech recognition (STT)
- AssemblyAI multilingual — 6 languages with real-time auto-detection (EN, ES, FR, DE, IT, PT)
- Deepgram Nova 3 — used for GCC region numbers; full Arabic support, 36+ other languages
Locking the language
By default the agent matches the caller. To force a specific language regardless of what the caller speaks, add this line to your system prompt:Backchannels and fillers
The engine speaks small filler words while listening — “mm-hmm”, “I see”, “right” — to make the call feel natural. These match the caller’s language automatically across all 15 supported languages.Numbers and pronunciation
Prices, phone numbers, dates, and addresses are normalised before being sent to the TTS. The agent says “twenty-five pounds” rather than spelling “two-five-pound-sign”. This works across languages without configuration. For brand or technical pronunciations the agent gets wrong, you can override them in the system prompt:What’s not supported yet
- Voice cloning per language (planned)
- Per-call language switching mid-conversation (the call locks to one language at start)
- Languages outside the 15 TTS list above