Why Decentralised AI
CENTRALISED AI COST
Frontier cloud LLM APIs charge per token. A LightNet Baby answers for free on your device, once compiled.
HALLUCINATION
LLMs frequently invent facts. A LightNet Baby uses deterministic lookup — within its own loaded knowledge set, every answer traces to a fact you wrote.
PRIVACY
Cloud AI queries are typically logged and may be used for training. A LightNet Baby answer never leaves your device.
OFFLINE
Cloud AI needs internet. LightNet Chat (working today) runs between two devices over Wi-Fi/Bluetooth with no connectivity.
FOOTPRINT
A specialist Baby compiles to ~35 KB native binary via the NovaGlyph Kit. Different problem class than a generative LLM — not a drop-in replacement.
ATTRIBUTION
Cloud LLMs can't tell you where a fact came from. Every Baby answer points back to the fact in the pack that produced it.
DISTRIBUTED KNOWLEDGE MESH

LIGHTNET

A mesh of specialist AIs running on personal devices — connected by Bluetooth — collectively providing knowledge without hallucination, without a central model, without sending your data anywhere.

The Concept Grid

Every Keyword Is a Concept.

Light LanguageLight LanguageVertexpace's 64-concept programming language. Every instruction maps to one documented keyword. Compiles to native machine code — the same bytes that run programs also encode mesh packets. is a compiled programming language where every keyword maps to a position in a documented compact concept encoding. Source code compiles to native machine codeNative machine codeInstructions that run directly on the CPU — no interpreter or runtime in between. Light Language compiles all the way down, producing ultra-compact binaries.— and the same concept bytes that run programs also encode mesh packets for Bluetooth transmission. Code and communication share the same alphabet. No hidden state, no translation overhead.
  • Compact encoding — the full instruction set fits in a single-byte alphabet.
  • Native compilation — lightweight binaries with zero runtime dependencies.
  • Dual encoding — program instructions and mesh wire packets share the same bytes.
-- Light Language source (illustrative)
ENABLE main()
    EXIST x BECOME 42
    SEND "LightNet alive"
END
→ watch the concept grid run live
Deterministic Inference

Babies That Reason.

Baby AIsBaby AIsCompact, private AI agents built by Vertexpace. Each one runs entirely on your hardware from a tiny compiled file — no cloud, no subscription, no data leaving your device. are compact knowledge bases compiled to native binariesNative binariesPrograms compiled directly for your hardware. No interpreter or runtime required — just download and run. Specialist AIs typically compile to 35–50 KB.. Each one knows a specific domain — first aid, electrical codes, field triage — and answers questions about it instantly, offline. When a Baby receives a word it has never seen, it follows the knowledge chain and knows exactly what to do. Meaning propagates through the network without a shared dictionary or a probabilistic modelProbabilistic modelAn AI that predicts the most likely next word based on training data — like GPT. Can produce fluent but invented facts. Baby AIs use deterministic rules instead..
  • Deterministic knowledge — rules compiled directly into the binary, no model required.
  • Inheritance resolution — related concepts chain automatically to the right answer.
  • Zero hallucination — every answer traces back to a specific fact you defined.
The Mesh Protocol

Bluetooth. No Internet. No Server.

Every device running LightNetLightNetVertexpace's mesh protocol for connecting specialist Baby AIs over Bluetooth. Devices share answers peer-to-peer — no server, no cloud, no central model required. advertises its specialist topics over Bluetooth Low EnergyBluetooth Low EnergyA short-range wireless standard designed for small data bursts with minimal battery use. LightNet uses it so devices find each other automatically within range — no Wi-Fi or pairing needed.. When you ask a question, your device broadcasts to the mesh. The specialist Baby that knows responds. Queries spread — Baby memories don't. Your personal data never leaves your device. Only queries and answers travel the mesh, attributed to the specialist that answered.
  • BLE discovery — devices find each other automatically within range.
  • Wi-Fi escalation — if no nearby peer knows, route over LAN or internet to trusted friends.
  • Answer attribution — every response names which Baby said it.
  • Privacy by architecture — queries are pseudonymous, memories never shared.
Compact Concept Encoding

Programs as Packets.

Because every Light Language keyword maps to a single concept byte, entire programs can be encoded as ultra-compact binary sequences. A complete specialist Baby — with facts and inference logic — fits inside a standard Bluetooth Low Energy packet. Send a whole AI over a radio burst. No serialisation, no JSON, no base64 — the program bytes ARE the transmission bytes. Devices on the mesh can receive intelligence, not just data.
  • Single-byte alphabet encodes both code and wire packets.
  • Standard BLE packet fits a full specialist Baby with inference logic.
  • Transparent — every byte decodes back to readable source. Trust is built in.
How a Specialist Baby compares to a cloud LLM

Different problem class.

AreaLightNet Specialist BabyCloud LLM
Footprint~35 KB native binary on your deviceHundreds of GB of weights running on server farms
HallucinationHallucinationWhen an AI invents a plausible-sounding fact that isn't true. LLMs hallucinate because they predict likely text, not verified facts. Baby AIs use deterministic rules — every answer traces to a fact you wrote.Within its loaded knowledge set, traceable to a fact you wroteProbabilistic generation; can invent facts
InternetNot required after compileAlways online
PrivacyQuery never leaves the deviceQueries typically logged; may be used for training
What it does wellOne narrow domain you scoped, deterministicallyOpen-ended generation, summarisation, code drafting
What it can't doAnything outside the facts you loadedGuarantee a specific fact came from a known source
Benchmarks run on Apple Silicon M2. Results vary by hardware and fact set size.
The Vision

More Devices. More Knowledge.

Every phone, laptop, and device runs its own tiny AI — each one specialised. One knows weather, one knows recipes, one tracks a medical condition. When you ask a question, your device broadcasts to every nearby device over Bluetooth. The specialist that actually knows responds. The more people who install it, the more topics get covered. The network gets smarter as it grows. Eventually: dedicated LightNet chips in keychains, cars, and appliances — pre-taught specialists that join the mesh automatically.
  • Specialist templates — first aid, weather, recipes, local news, agriculture, language translation.
  • Community-taught — anyone can build a specialist Baby for their topic.
  • Hardware roadmap — dedicated LightNet chips for always-on mesh presence.

GET IN TOUCH.

Interested in LightNet, Lumina OS, or any other VertexPace project? Send me an email and I'll personally help you get set up with the right build for your hardware.
EMAIL ZAC
robertpage@vertexpaceorganization.com
© 2026 VERTEXPACE. NOVAGLYPH KIT AND BABY AI LITE ARE PROPRIETARY PRODUCTS.
LIGHT LANGUAGE SPECIFICATION AND CORE COMPILER SOURCE AVAILABLE UNDER OPEN-SOURCE TERMS.
STATUS: ACTIVE R&D · SOVEREIGN BY DESIGN · 🐝