We name the parts we buy.
We design each agent and its call flow ourselves. Outside providers are named product by product, once verified.
- Voice-activity detection
- Named per product · Sori AI: Silero VAD
- Outside providers
- Named per product once verified · Sori AI: NAVER CLOVA Speech · NAVER CLOVA Studio · NAVER CLOVA Voice · LiveKit Cloud
As of
How to read a call trace.
Each answer the agent gives is one bar: what came in above, what went out below. Drawn from a synthetic script, not a recording.
- 1Upper run · what came in
- Speech-to-text writes the words; the model decides.
- 2Lower run · what went out
- Tools do the work; speech-out says the reply.
- 3Hub 2 × 4 · turn in flight
- Blue only while a live turn runs.
- 4Length · scale
- A script has no timings; lengths are fixed.0100 ms1 cell = 10 ms · once real logs are connected
We publish latency only once we have measured it: p50 and p95 per language, each with its date.
Products deploy three ways.
Depending on the product, it runs in the cloud, on your servers, or split between them.
- CloudEverything runs outside; call audio, transcripts and tool calls leave your premises
- On-premisesThe whole runtime runs on your servers, where the product supports it; only outside tools receive data
- HybridYou choose which stages stay inside; the rest run in the cloud
What runs where, model by model
| Model | Runs on your servers | Runs outside | What leaves your premises |
|---|---|---|---|
| Cloud | Nothing. Only your phone line and the systems we connect to stay with you. | Speech-to-text, the model and its tool calls, speech-out, call logs | Call audio, transcripts, tool-call contents |
| On-premises | The whole runtime: speech-to-text, the model and its tool calls, speech-out, call logs | Only tools you connect to outside services, such as SMS | Only what those outside tools are sent |
| Hybrid | The stages you choose to keep inside | The rest | Whatever the outside stages process |
Which models a product supports varies, and each product page says so. Dallang, for example, ships as SaaS or on-premises. For hybrid, we decide together in the working session which stages stay inside, against your security rules.
Systems the tool calls write to
- POS
- The priced order · Orda
- PMS
- Guest requests, sorted by department (your PMS connection is reviewed in the working session) · gohotel.io
- CRM
- The caller and what they asked for (Overture AI writes to its built-in CRM) · Overture AI
- Calendar
- The booked slot, in Sori AI's own booking calendar · Sori AI
We confirm the product and version of each system you run in the working session.
How latency will be published
- Metric
- p50 and p95, in milliseconds
- Split by
- Language and provider
- Measured on
- Our internal test bench
- Published with
- Date · calls measured · line type · server location
- Broken down by
- Speech-to-text · model · tools · speech-out, as in the trace
Bring your security rules to a working session.
Tell us the deployment model, the call languages and the systems to connect, and we send an agenda built around them.
Book a working session