BUILT ON GOOGLE CLOUD VERTEX AI & GEMINI

Sub-300ms multimodal voice & avatar agents for production apps.

KurisuStarts AI provides developer APIs and WebSocket SDKs that turn Google Gemini models into expressive, real-time voice and 2D/3D visual agents—with speculative sentence streaming, emotional prosody, and instant barge-in.

▶ Launch Live Amadeus MVP Demo Read Architecture Spec

185ms

First-Sentence Audio TTFB

60 FPS

3-Band Formant Lip-Sync

99.94%

Cloud Run & Vertex Uptime

amadeus-v2.kurisustarts.com // wss://stream VERTEX AI ONLINE

Amadeus Neural Agent (v2.4)

                    Model: gemini-3.5-flash • Emotion: analytical (0.88)
                  

Explain how KurisuStarts reduces voice agent latency below 300ms.

Instead of waiting for the full LLM completion, our Vertex AI streaming pipeline extracts the first clause at token 8 and dispatches it immediately to our FP16 neural acoustic synthesizer while streaming the remaining clauses asynchronously over WebSockets.

TTFB: 182ms (Speculative Split) Vowel Formant: /a/ • /i/ • /o/ synced

PROPRIETARY MULTIMODAL ENGINE

Why developers replace stitched TTS wrappers with KurisuStarts.

Traditional voice agents chain separate STT, LLM, and TTS REST calls—creating 2.5+ second awkward pauses. KurisuStarts unifies speculative Gemini streaming, emotional acoustic conditioning, and client-side WebAudio formant analysis in a single pipeline.

01 // SPECULATIVE STREAMING

Clause-1 Speculative Extraction

Our streaming orchestrator parses structured emotion metadata and first-sentence boundaries mid-stream from Google Vertex AI Gemini models—triggering neural audio playback in under 200ms while subsequent sentences synthesize in parallel.

02 // ACOUSTIC LIP-SYNC

3-Band Formant Vowel Rigging

Instead of crude volume-based jaw movement, our client-side WebAudio analyzer isolates F1/F2 vocal tract frequencies (300–900Hz & 900–2800Hz) to drive real-time /a/, /i/, /u/, /e/, /o/ mouth shapes at 60 FPS on HTML5 Canvas and WebGL.

03 // EPISODIC MEMORY

Hippocampal Engram Retrieval

Persistent user facts, temporal schedules, and emotional divergence scores are indexed automatically into low-latency vector & relational memory—giving agents long-term continuity across web, mobile PWA, and telephony sessions.

CLOUD-NATIVE INFRASTRUCTURE

Engineered natively on Google Cloud & Vertex AI.

KurisuStarts AI is architected from day zero on Google Cloud’s AI-native stack to deliver deterministic low latency, enterprise data isolation, and global autoscaling.

Our Google Cloud Production Stack

Vertex AI (Gemini 3.5 / 4 Flash & Pro): Primary reasoning, multimodal vision inspection (screen/webcam PiP), and structured emotional state JSON generation.
Google Cloud Run & GKE: Autoscaled WebSocket gateway (`wss://`) handling concurrent bidirectional audio & barge-in control frames.
Compute Engine L4 / T4 GPU Nodes: FP16 neural voice synthesis (GPT-SoVITS v2 & custom acoustic prosody filters) with zero-copy OGG Vorbis streaming.
Cloud Storage & BigQuery: Voice model weight distribution, latency telemetry warehousing, and enterprise audit logs.

// Quickstart: Initialize a real-time KurisuStarts Multimodal Agent
import { KurisuAgentClient } from "@kurisustarts/sdk";

const agent = new KurisuAgentClient({
  apiKey: process.env.KURISUSTARTS_API_KEY,
  region: "us-central1", // Google Cloud Vertex AI region
  backbone: "vertex-gemini-3.5-flash",
  voiceProfile: "amadeus-neural-fp16",
  lipSyncMode: "formant-3band-60fps",
});

await agent.connectWebSocket({
  onEmotionPreview: (state) => rig.setExpression(state.emotion),
  onAudioChunk: (oggBuffer, phonemes) => audioEngine.enqueue(oggBuffer),
});
          

BUSINESS MODEL & TRANSPARENT PRICING

Usage-based SaaS pricing built to scale from prototype to enterprise.

Start prototyping with our free Builder tier, or scale to millions of real-time conversational minutes with dedicated Vertex AI throughput.

BUILDER / MVP

Starter

$0 / month

For solo developers validating multimodal voice and avatar workflows.

150 real-time voice + avatar minutes / mo
Vertex AI Gemini Flash reasoning backbone
Standard 60 FPS WebCanvas avatar rig
WebSocket & REST API access
Community Discord & docs support

Request API Key

Growth SaaS

$79 / month

For SaaS products, AI companions, and interactive customer agents in production.

2,500 real-time voice + avatar minutes / mo ($0.028/min overage)
Speculative Clause-1 sub-200ms streaming
3 custom neural voice clones (FP16 GPU)
Persistent Hippocampal Memory API
Multimodal Webcam & Screen Vision PiP

Start 14-Day Free Trial

DEDICATED CLOUD

Enterprise

$399 / month

For high-volume teams requiring dedicated Google Cloud GPU & Vertex AI quotas.

15,000+ included minutes & volume discounts
Dedicated Vertex AI Provisioned Throughput
Custom 2D vector or Live2D/WebGL rig integration
Regional data residency on Google Cloud
99.95% SLA & dedicated founder engineering support

Contact Founder & Sales

ABOUT KURISUSTARTS AI

Building the human interface layer for frontier AI.

Founded in 2026 by AI systems engineer Suleman Ali Butt, KurisuStarts AI grew out of Project K.R.I.S.T.I.N.A. (Amadeus 2.0)—an engineering effort to eliminate the robotic latency barrier between large language models and expressive human communication.

Our live production MVP is deployed and publicly accessible today, combining Google Cloud’s Gemini models with real-time neural acoustics, visual perception, and long-term episodic memory. We are currently scaling our self-serve developer platform and preparing for our upcoming pre-seed institutional round.

Inspect Live Production App ↗ founder@kurisustarts.com

CORPORATE & TECHNICAL PROFILE

Company Fact Sheet

Company Name	KurisuStarts AI
Primary Domain	kurisustarts.com
Founder & CEO	Suleman Ali Butt
Founder Contact	founder@kurisustarts.com
Founded	2026
Primary Cloud Stack	Google Cloud (Vertex AI, Cloud Run, GKE)
Product Stage	Live Production MVP (v2.4)