AI features
Where the voice agent actually sits
Quoral is in the business of building a core AI/ML product. These named jobs are first-class on the line: the agent runtime, the models it can call, and the telephony that carries them. This is a voice agent company — not a chat widget bolted onto a number.
AI feature · Voice agent API
Inbound and outbound phone agents on one agent id
For teams that need a spoken agent on PSTN, SIP, or a WebSocket
Write flows, tools, and knowledge once. Quoral runs the agent on the live line. Studio and the API share the same agent id so a patch in the booth ships to production.
Flow + tools + knowledge → POST /v1/runs → live call
AI feature · Sub-second turn-taking
Barge-in, silence, and the next line under a second
For operators who cannot wait on a contact-center pause
Quoral owns turn-taking on the audio path. The caller can interrupt. Silence does not hang. Every spoken turn is designed to stay under a second.
PCM / RTP → turn-taking runtime → next utterance
AI feature · Model routing
Claude, GPT-4o, Llama 3 70B, and Amazon Nova
For platform teams picking speed on the spoken turn
Planned model families: Claude 3.7 Sonnet, GPT-4o, Llama 3 70B, Amazon Nova. Access paths: Bedrock, Anthropic, OpenAI Direct, Groq. Agent frameworks: custom/homebrew and Amazon Bedrock AgentCore. Quoral is not fine-tuning those models.
Agent turn → selected model → Bedrock / Anthropic / OpenAI / Groq
AI feature · Tools, RAG, transcripts
Function calls, site-synced knowledge, one call object
For builders who need the agent to book, look up, and hand off
Tool calls stream while the caller is still speaking. RAG syncs from your site onto the same agent. Transcripts, traces, and recordings land in one call object you can score.
Utterance → tool / RAG → transcript on the call object
IVA platform. Speech recognition, chat, and customer-support voice. MVP stage — no invented customers, funding, or uptime claims on this page. Models are consumed, not trained here.