wubbery://engine — a CPU-native execution substrate
hundreds of measured modules beneath the operating system. 85 of them are things that don’t happen.
Memory placement, power allocation, capacity planning, inference admission, agent governance — each is somebody’s whole company, and none of them shared a substrate until now. Most of what it does is refuse: a rack that would trip on failover, a compute purchase that returns nothing, a generation that would die at token 400,000. Refusals are hard to fake and easy to demonstrate, which is why we count them instead of averaging them.
The primitive underneath is a decision costing 3.67 µs on one core — no GPU, no network hop, no model call. That is the reason it can sit in front of everything; it is not the product.
WUBBERY Engine vs. Legacy Cloud AI Stack
Measured on an 11th-gen i5-1135G7 laptop, single-threaded, best-of-5 over 100,000 decisions. Re-run any row yourself: GET /v1/bench/overhead.
| Engine Capability | ⚡ WUBBERY Substrate Engine | Legacy Cloud AI SaaS Stack | Enterprise Business Impact |
|---|---|---|---|
Canonical Decision Latency Category: throughput | 3.67 μs (0.00367 ms) | 1,800 ms - 2,400 ms | 27,248x FASTER THAN A 100ms MODEL CALL IN THE PATH |
Single-Core Decision Volume Category: throughput | 272,307 Decisions / sec | 0.4 Decisions / sec | HIGH-DENSITY RACK THROUGHPUT |
Realtime Viseme Mouth Alignment Category: throughput | < 15 ms (60 FPS Lip-Sync) | 750 ms - 2,200 ms Lag | ZERO SPEECH-TO-TEXT BUFFERING |
Per-1k Turn Token Royalty Category: cost | Price delta between tiers | $12.00 - $120.00 / 1k Turns | 32.9% MEASURED COST REDUCTION |
Enterprise Data Centre Cost Category: cost | 32.9% measured on real replayed traffic | Every request at the top tier | 69% VS AN ALWAYS-PREMIUM BASELINE |
Facility Energy & Power Draw Category: esg | Inference energy falls with the routed share | All inference at premium tier | DEPENDS ON YOUR AI SHARE OF IT LOAD |
Statutory ESG Filing Compliance Category: esg | Local audit log of every routing decision | Provider-side logs you do not hold | APPEND-ONLY JSONL, TIER LABELS ONLY |
On-Device Memory Encryption Category: security | AES-GCM 256; keys never leave the process | Cloud Gateway Telemetry Logs | LOOPBACK-ONLY BIND, NO EGRESS |
Algorithm Code Protection Category: security | Server-only core; never shipped to a client | Exposed API Keys & Plaintext | COMPILED, NOT OBFUSCATED — SEE /SECURITY |
Multi-Platform Delivery Formats Category: delivery | Anthropic + OpenAI dialect gateway; Docker | Restricted SaaS HTTP Endpoint | RUNS IN YOUR VPC, BINDS LOOPBACK |
Executable Terminal Daemon Category: delivery | Single Node process, one env var | Complex Cloud Setup | ANTHROPIC_BASE_URL=127.0.0.1:8787 |
Facility Scenarios & WUBBERY Substrate Impact
Pick a facility size to see what the measured reductions come to at that scale. These are scenarios, not customers — the sizings are illustrative, and the savings are our benchmarked ratios applied to them.
Complete Substrate Modular Architecture
The engine's layers. Routing, memory and prefetch are live and benchmarked; the media layers are roadmap and labelled as such.
@wubbery/engine-video60 FPS (16.6ms frame budget)Realtime Generative Video & Motion Layer
Generative Video (roadmap)Roadmap. Not shipping today — there is no released package for it, and the site should not imply otherwise.
@wubbery/engine-viseme< 15ms Latency (60 FPS Visemes)Sub-15ms Acoustic Viseme Classification Layer
Real-Time Web Audio FFT EnginePerforms real-time frequency domain (FFT) mouth shape classification from audio streams directly inside the browser audio worklet.
@wubbery/engine-voice< 45ms Local Audio SynthesisClient-Side Voice Imprint & TTS Synthesis Layer
Neural Voice (roadmap)Synthesizes natural neural speech from text using lightweight ONNX models running inside browser WebWorkers.
@wubbery/engine-router3.67 μs (272,307 Decisions/sec)Governed Intent Routing & Model Ensemble Layer
Substrate Autonomy Router (9 Lanes)Routes prompts dynamically across whichever frontier models you already use, based on cost, latency, and permission boundaries.
@wubbery/engine-vault< 4ms AES-GCM EncryptionSealed Encrypted Memory & Patent Enclave Layer
Harkin Theorem Sealed Security VaultEnforces strict on-device memory encryption, sealing Harkin Theorem patents, private memory keys, and enterprise customer data.
wubbery/engine-sidecar< 28ms IPC GatewayNative Runtime & Audit Ledger Layer
Native C++ / Python Runtime DaemonHigh-density native runtime daemon for enterprise data centres and desktop operating systems with cryptographic audit logging.
Put In Your Figures. See What The Measured Ratios Do To Them.
We do not know what you spend, so we do not guess. You supply the baseline; we supply the reductions we have measured, and every line below names the benchmark it came from.
Typical corporate facility overhead means every watt of compute removed takes roughly half a watt of cooling and distribution with it.
Picking an industry only changes the starting numbers below and the assumed facility overhead. The measured reductions are the same for every industry on this list — they are a property of the engine, not of who is running it.
Annual spend on inference that could be routed. Not your whole cloud bill.
Servers in your own facility running this workload.
Your average draw per server. Ours is not a substitute for yours.
Your contracted electricity price.
Facility overhead (PUE): total site draw ÷ compute draw. 1.0 shows compute alone. Yours, not ours.
$14.00M / yr
Adding the levers instead of composing them on the remainder gives 116.64% — more than the spend itself. We publish the gap rather than the bigger number.
- These are estimates from your inputs, not a measurement of your systems.
- The ratios are measured on our benchmark traffic. Your mix is not our mix — the shadow replay runs against your own logs and returns your number, without changing what you serve.
- Compute energy is compute draw only. Cooling, PUE and your AI share of IT load are inputs only you have.
- Cost avoided and capacity gained have different denominators, so they are never combined into a single percentage. They are added once, as money, only to price the share — and both halves stay on screen so the addition can be checked.
- Capacity is a modelled figure: the compute saving is measured in relative compute units against a declared model, not in tokens per second on hardware.
GET /api/wubbery/proof — no key, no input, computed on request.414 models. Pick any of them to compare with vs. without WUBBERY.
The full routable catalogue across 51 providers, pulled from the OpenRouter public model API on 2026-08-20 — prices and context windows are theirs, not ours, and not retyped. None of these carry a WUBBERY benchmark yet: pick the ones you actually run and we’ll measure those.
Enterprise Financial & Energy ROI Engine
Enter your exact data centre server rack count, monthly API volume, and energy costs to compute exact mathematical savings.
Ask WUBBERY Engine Assistant
Inquire about Hyperscale AI facility's routable spend, the memory index, or how to run the benchmarks yourself.
Hello! I am the WUBBERY Substrate Engine Assistant. How can I help you analyze Hyperscale AI facility (ARCHETYPE — NOT A REAL ORGANISATION) or test our 3.67μs intent routing layer?
Deliver WUBBERY Engine Modules Anywhere
Integrate standalone WUBBERY modules into your web apps, mobile builds, or enterprise on-prem containers.
Generative Video Canvas Engine
Renders dynamic lip morphing, breathing sway, and facial deformation on native HTML5 Canvas & WebGL without sending video frames over the wire.
import { ArteiGenerativeEngine } from "@wubbery/engine-video";
const engine = new ArteiGenerativeEngine(canvasElement);
engine.updateViseme({ mouthOpen: 0.85, mouthWidth: 0.4 }, { fps: 60 });Request Substrate Engine Enterprise Access
Create your enterprise profile and verify your business email to unlock sub-4μs intent routing access.