This is ZeroRuntime's home on Hugging Face for the models we build in house and the research behind them. Everything here is designed to give developers and researchers production-grade starting points for real-time voice, backed by the same work that powers the ZeroRuntime platform.
What is ZeroRuntime?
An AI voice agent infrastructure platform for building, deploying and running production voice AI agents across web, mobile, telephony and physical devices.
- One optimized system. Transport layer, agent logic and voice components together, so orchestration overhead and latency both drop.
- Agents start on demand. No pre-reserving, so concurrent sessions scale without you managing the infrastructure.
- Your stack, your choice. Bring your own LLM, STT and TTS.
| Runtime | Speech, models and audio in one system. Up to 70% lower latency |
| SDKs | Voice agents for web, mobile, telephony and devices |
| Cloud | Managed infrastructure, agents start on demand |
| Playground | Test live, switch models, inspect latency, tune prompts |
Research
We build proprietary models when off-the-shelf solutions fall short on speed or accuracy.
Echo: turn detection
Knowing when a person has actually finished talking is the difference between a conversation and an interruption. Silence timers guess. Echo does not.
| Model | Reads | Built for |
|---|---|---|
| Echo Small | Transcript | The lowest latency |
| Echo Large | Transcript | Higher accuracy |
| Echo Omni | Audio + transcript | Multimodal accuracy across 27 languages |
- Up to 97.3% completed-turn recall at over 93% overall accuracy
- Four states, not two: Complete, Incomplete, Backchannel, Wait
- Trained on data we build in house, for a problem public corpora do not cover
Enterprise
- Encryption, audit trails and certifications that clear security and procurement
- Your data stays yours. No third party touches a call
Your next voice AI experience starts here.
Docs · Code samples · Pricing · Blog · Community
San Francisco · Surat