Review configuration compatibility
Wirevox is capability-aware. A control can be executed natively by a provider, enforced by the Wirevox pipeline, added to the model as guidance, or inactive for the selected architecture.
Support terms
| Term | Meaning |
|---|---|
| Native | Sent to and executed by the selected provider. |
| Wirevox | Enforced by a Wirevox runtime component. |
| Guided | Added to model or voice instructions; behavior should be tested because it is not a hard provider control. |
| Inactive | Preserved where applicable but not executed by the selected pipeline. |
Architecture-level controls
| Control | Modular | Realtime |
|---|---|---|
| System prompt | Wirevox prompt compiler | Wirevox prompt compiler |
| Custom greeting | Wirevox call start | Wirevox call start |
| Primary language | Wirevox speech session | Wirevox speech session and provider context |
| Primary voice | Native speech provider | Native realtime provider |
| Speaking speed | Native where supported; otherwise guided | Guided through realtime voice instructions |
| Silence prompt timeout | Wirevox lifecycle | Provider- or Wirevox-managed by provider |
| Auto-hangup timeout | Wirevox lifecycle | Wirevox lifecycle |
| Maximum call duration | Wirevox lifecycle | Wirevox lifecycle |
| Knowledge base | Wirevox retrieval | Wirevox retrieval |
| RAG threshold | Wirevox retrieval | Wirevox retrieval |
| Post-call webhook | Wirevox post-call processing | Wirevox post-call processing |
| Audio environment | Wirevox media output | Wirevox media output |
Controls that differ by architecture
| Control | Modular | Realtime |
|---|---|---|
| STT provider | Native Modular STT selection | Inactive; provider owns recognition |
| STT patience / response speed | Native Modular endpointing | Inactive; use Realtime turn-taking |
| Interruption sensitivity in words | Wirevox barge-in threshold | Inactive; provider emits interruption events |
| Filler words to ignore | Wirevox barge-in filtering | Inactive |
| Secondary language | Wirevox language routing | Inactive |
| Secondary voice | Wirevox TTS routing | Inactive |
| Responsive / Balanced / Patient turn-taking | Inactive; use STT patience | Native provider VAD mapping |
Realtime provider differences
| Control | OpenAI Realtime | Grok Realtime | Gemini Live |
|---|---|---|---|
| Silence prompt timeout | Native | Native | Wirevox lifecycle |
| Keyterms | Native transcription context | Native transcription configuration | Guided in session instructions |
| Pronunciation replacements | Guided in session instructions | Native session setting | Guided in session instructions |
| Reasoning effort control | Inactive | Native (none or high) | Inactive |
| Turn-taking preset | Native provider VAD | Native provider VAD | Native provider VAD |
The table describes the current Wirevox implementation, not every feature a provider may expose in its own API.
Voice compatibility
In Modular mode, the LLM and TTS provider can be selected independently from the compatible voice catalog.
In Realtime mode, the voice must belong to the selected realtime provider:
- OpenAI Realtime uses an OpenAI-compatible realtime voice;
- Grok Realtime uses an xAI-compatible realtime voice; and
- Gemini Live uses a Gemini-compatible realtime voice.
When a Realtime model change also changes provider, the editor selects that provider's default voice. Listen to the result and explicitly choose another compatible voice if necessary.
Modular speech-recognition compatibility
The current Modular editor can offer:
- Deepgram Nova-3;
- Deepgram Flux Multilingual;
- Gladia; and
- Cartesia Ink 2 for an English-only configuration without a secondary language.
OpenAI Whisper can appear as Soon and is currently disabled in this selector.
The editor can replace or disable an incompatible STT choice when the primary or secondary language changes.
Understand editor diagnostics
Compatibility messages can be informational or blocking. Current examples include:
- secondary-language settings are inactive in Realtime;
- Modular STT patience is inactive in Realtime;
- word interruption and filler filtering are inactive in Realtime;
- Realtime turn-taking presets are inactive in Modular; and
- a registered Coming Soon model is not available.
Do not copy a diagnostic into the system prompt. Change the architecture or configuration if the behavior is required.
Verification checklist
After changing architecture, model, provider, language, or voice:
- Reopen all affected configuration sections.
- Confirm the active model and voice.
- Check secondary-language and STT availability.
- Review Voice Delivery and Conversation Settings.
- Resolve blocking diagnostics and understand informational ones.
- Run an interruption, silence, pronunciation, and action test.
- Check the draft cost breakdown.
- Publish before expecting live calls to use the new configuration.