Vozzo provides two powerful AI model architectures for voice agents. Choosing the right model directly affects latency, naturalness, and capability.
Model Overview
Orbit
vz_model_id: 1Ultra-low latency Speech-to-Speech (STS) model. The entire voice pipeline runs in a single unified neural model — no separate STT or LLM step.
- Pipeline: STS only
- Latency: ~200 ms
- Best for: Real-time, low-latency calls
Quantum
vz_model_id: 2A modular three-stage pipeline: Speech-to-Text → LLM → Text-to-Speech. Each component is independently configurable with your preferred provider.
- Pipeline: STT → LLM → TTS
- Latency: ~600–900 ms
- Best for: Complex reasoning & customisation
Pipeline Comparison
| Feature | Orbit | Quantum |
|---|---|---|
| STT Provider | Unified | Configurable (Google, Deepgram, etc.) |
| LLM | Built-in | Configurable (GPT-4, Claude, Gemini, etc.) |
| TTS Provider | Unified | Configurable (ElevenLabs, Cartesia, etc.) |
| Latency | ~200 ms | ~600–900 ms |
| Custom Prompts | ✓ | ✓ |
| Knowledge Base | ✓ | ✓ |
| Function Calling | Limited | Full support |
| Fallback Config | ✗ | ✓ |
Setting the Model via API
Pass vz_model_id when creating or updating an agent. Use the Authorization: Token prefix for API keys.
cURL
| Value | Model |
|---|---|
1 | Orbit |
2 | Quantum |
Retrieve Available Models
Fetch the current model list from the master data endpoint.
cURL
Code
Model Config (Quantum Only)
For the Quantum model, each pipeline stage is individually configurable via agent_model_config:
Code
config_id | Stage |
|---|---|
| 1 | STT (Speech-to-Text) |
| 2 | LLM (Language Model) |
| 3 | TTS (Text-to-Speech) |
| 4 | STS (Speech-to-Speech, Orbit only) |
Fetch available providers and models per stage using the config master endpoint.
cURL