Now in private beta

VoiceLayer

Build voice AI layer by layer, ship in hours.

Ten production primitives for voice agents, from telephony and speech runtime to memory, workflow, and security.

Request early access How it works

<800msP90 latency

10Primitives

1 APIOne surface

Your Voice Agent

Primitives 07-10

Trust and observabilityTranscripts, events, handoff, and structural security wrap the stack.

Primitives 01-06

Voice, AI, and workflowSpeech runtime, memory, and process logic stay composable from the first call.

Agent Control Plane and Runtime

Layers 6-9

Agent execution and memoryIntelligence lives inside the call path, with the upper stack carrying execution, orchestration, and state.

Layers 4-5

Programmable identity and complianceA middle control band keeps policy, verification, and state transitions pinned to the same runtime path.

Layers 1-3

Owned networkThe foundation layer carries routing, delivery, and transport so the upper bands can separate without losing continuity.

Layers 6-9

Agent execution and memory

Layers 4-5

Programmable identity and compliance

Layers 1-3

Owned network

Platform

Three system layers, ten deployable primitives.

The old HTML had two separate primitive walkthroughs. This keeps one: a grouped summary up top and a single detailed list below.

Primitives 01-03

Voice and speech

Telephony, audio intelligence, and the streaming STT -> LLM -> TTS runtime that gets first audio back fast.

Voice Infrastructure
Audio Experience
AI Speech Runtime

Primitives 04-06

AI and workflow

Memory, intent classification, and typed process logic that keep each call stateful and deterministic.

Memory Engine
Intent and Signal
Process Schema

Primitives 07-10

Trust and observability

Transcripts, handoff, routing, and structural security wrapping every interaction in production controls.

Transcript Intelligence
Handoff Engine
Event and Routing
Security Primitive

Voice Infrastructure

SIP trunking, WebRTC, number provisioning, and tenant-aware dispatch in one surface.

SIPWebRTCRouting

Audio Experience

Language detection, noise handling, audio profiles, and DTMF fallback before reasoning starts.

DetectionProfilesFallback

AI Speech Runtime

Overlapped STT, LLM, and TTS optimized for sub-second first response times.

StreamingRealtimeLatency

Memory Engine

Caller history, account context, and multi-scope memory pre-warmed during ring time.

ContextRecallPre-warm

Intent and Signal

Parallel classification for intent, churn risk, escalation cues, and business actions.

IntentRiskSignals

Process Schema

Typed workflows for collect, branch, confirm, tool call, and handoff behavior.

DSLTypedWorkflow

Transcript Intelligence

Live transcript ingestion with async enrichment, summaries, and searchable records.

TranscriptSearchEnrichment

Handoff Engine

Warm transfer with preserved context, queue routing, and continuity for the caller.

TransferQueuesContext

Event and Routing

Hot-path, side-channel, and cold-path events without contaminating audio latency.

EventsBullMQAsync

Security Primitive

Input, prompt, output, and action guards that remain on for every workflow.

GuardrailsPIIPolicy

How it works

One call path. Hot path discipline. Cold path everywhere else.

The platform is designed around sub-second response time on the live call and deferred work off the audio path.

Call arrives and agent dispatch resolves the tenant

Inbound traffic lands on the voice edge, the room is created, and the right worker is assigned.

<500ms boot

Memory is pre-warmed while the phone is still ringing

Caller identity, history, language, and business context are loaded before first utterance.

~5ms hidden

Speech, signals, and workflow checks execute in parallel

Audio flows through STT while intent, process state, and safety rules evaluate concurrently.

~50ms logic

LLM streams and TTS starts speaking on the first useful token

The response is generated progressively rather than waiting for a full completion.

~400ms overlap

Cold-path events fire without touching the audio path

Transcripts, CRM writes, memory updates, and downstream events continue after the caller hears the answer.

0ms added to hot path

Use cases

Designed for production voice systems, not demos.

Use the same stack for support, qualification, intake, collections, and any flow that needs memory, workflow, and guardrails together.

Inbound support

Answer, classify, resolve, and escalate with context already loaded before the first exchange.

24/7coverage

Sales qualification

Run scripted qualification flows, route by signal, and preserve transcript intelligence for follow-up.

<800msp90 first reply

Ops and collections

Automate structured calls that need deterministic workflow, auditability, and transfer safety.

10composable primitives

Early access

Use `vlayers.ai` as the public front door, keep the control plane behind auth.

The landing page is public. The existing Next.js dashboard remains at `/dashboard`, protected by Clerk.

Request access Open control plane