Kueizen

K1V4 — Your AI Companion

An AI that lives on your PC. No cloud. No subscription. No data leaves your machine.

An AI ensemble that auto-adapts to your hardware. It sees your screen, speaks in its own voice, remembers what matters, and gets better every generation. No prompt engineering. No configuration. Everything runs on your machine.

K1V4 vision — identifying images from your screen

Ask it about anything on your screen

K1V4 reading and outlining a PDF document

Feed it documents, get summaries and outlines

Analyze videos and web content

K1V4 evolution screen — genetic optimization

Evolves through genetic optimization

Sees

A dedicated vision model processes your screenshots, images, PDFs, and video content. The ensemble routes each query to the right model automatically. Ask about anything on your screen.

Speaks

Custom EVO VITS voice model trained on K1V4's own voice. 48kHz synthesis, phoneme-level lip sync on the avatar. Talk to it like a person — it talks back.

Remembers

Knowledge graph with episodic memory — not a chat log. Facts have confidence scores, temporal validity, and decay over time. It knows what matters and forgets what doesn't.

Acts

Web search, file operations, chess, math — native tool calling, not API wrappers. The AI decides what to use, executes it, and weaves the result into conversation.

Evolves

K1V4 is a GOLEM. Her intelligence is genetically optimized across hundreds of test scenarios. On our internal rubric, she scores 95.6% of the accuracy of a 100B+ parameter model — using 20x fewer parameters. She gets better every day, without you touching a thing.

Expresses

A fully animated VTuber avatar with eye tracking, lip sync, hair physics, and hours of hand-drawn expression animations. It reacts to what you're talking about — not canned loops.

Stays Local

AES-256-GCM encryption bound to your machine. 63+ prompt injection patterns blocked. Your conversations are encrypted at rest — even we can't read them.

System Requirements

Minimum

GTX 1060 6GB / 16GB RAM / Windows 10

Recommended

RTX 3090 / 32GB RAM / Windows 10

Adapts

Auto-detects your hardware — dual GPU, single GPU, or CPU-only. No configuration needed.

Early Access — Coming Soon on Steam

Wishlist on Steam

Add to your wishlist to get notified at launch.

Under the hood

Inference AI ensemble — multiple specialized models, auto-routed per query, adapts to your hardware

Voice Whisper STT (5 model sizes) + EVO VITS TTS (48kHz) + phoneme-to-viseme lip sync

Security Argon2id KDF → AES-256-GCM, machine-bound keys, 63+ injection patterns blocked

Hardware Auto-detects your GPU setup — 7 modes from dual-GPU to CPU-only, zero configuration

Engine Godot + Rust

For Teams & Enterprise

GOLEM

Self-Modifying Agent

A graph execution engine with hundreds of node types. Most agent frameworks chain LLM calls. GOLEM routes the majority of traffic through reflex layers that never touch an LLM — same quality, orders of magnitude cheaper.

Routing Evolved neural routers resolve routine queries at sub-millisecond latency — expensive models only fire when needed

Substrate LLMs, tools, neural networks, cellular automata, embeddings — composed into arbitrary graph topologies

Execution Streaming, parallel branches, failure policies, graph validation — production-grade, not a research prototype

ELMER

AI Deployment Infrastructure

Enhanced Language Model Execution Runtime. One-click mesh cluster for AI inference — deploy models across your hardware, route requests intelligently across providers, scale without re-architecting. ELMER automatically configures model parameters, provisions and benchmarks models, and runs ensemble pipelines. Enterprise-ready infrastructure you can stand up in minutes, not months.

Local Execute models locally with automatic parameter configuration

Catalog Extensively benchmarked model catalog — select, provision, and deploy

Ensemble Run ensemble models — provision, benchmark, and mix them in a single runtime

MOTHERSHIP

Evolutionary Core

300,000 lines of Rust built over a decade of independent research. Mothership evolves LLM prompts and graph architectures using Phantom Floor Search — a proprietary algorithm with hierarchical safety guarantees. Solutions improve across generations but can never regress below a known-good state. The optimization process itself gets smarter over time.

PFS A new class of evolutionary optimization not found in published literature — safety-first by design, not by constraint

Evaluation Multi-model jury system — multiple LLMs score independently. Implicit fitness from user behavior, no explicit ratings needed.

Scale 10 years of compounded research — Complex Systems, ALife, evolutionary computation. Battle-tested, not theoretical.

Deployment Cloud, on-prem, hybrid. Your choice.

Sovereignty Runs on your hardware, no external dependencies

Compliance FedRAMP, HIPAA, SOC 2 compatible — fully on-prem, no data egress

How It Works

Define fitness. Evolution does the rest.

Describe what "good" looks like — or let the framework generate the scenarios itself. Mothership breeds candidate solutions across generations, surgically fixing failures along the way. Safety floors guarantee you never regress. The best candidate deploys as a production GOLEM.

Many queries never hit an LLM.

GOLEM's reflex layers — evolved neural routers and embedding classifiers — handle routine traffic at sub-millisecond latency. Only novel or complex queries reach expensive models. Same quality, fraction of the cost.

Evolution that can't go backwards.

Phantom Floor Search guarantees that every generation is at least as good as the last. Safety floors, regression canaries, and hierarchical trust regions mean your production system never degrades — even while it's being optimized.

On-Prem Full Deployment

Zero Data Collection

Any GPU Consumer to Enterprise

No Lock-in Keeps Running if You Leave

Community

Follow development, ask questions, share feedback.

K1V4 Discord X / Twitter

Business

Enterprise deployments, partnerships, integration.

[email protected]

We build AIthat builds itself.