Private Mac apps you download today. Offline specialist tools that work without internet. And custom builds commissioned directly from Zac. One price, no subscription.
● Live · Mac + Windows
MAC APP
$29
FIX_AI
Fix AI
Offline computer diagnostic assistant for Mac and Windows. Scans your machine before you type — disk, memory, crashes, network — then diagnoses and fixes. LuminaVM triage engine. Black box recorder. Morning scan notifications. No cloud, no account.
Private AI chat app that runs entirely on your Mac. No cloud, no account, no logs. Powered by Ollama — bring your own models. Apple Silicon and Intel, macOS 13+.
An AI operating system that runs on your Mac and bare-metal on a Raspberry Pi 5. Every process is an AI agent. Every instruction is a plain English word from a 75-word vocabulary. Faulted agents repair themselves via a local LLM — no cloud, no subscription, no shared state.
34 working template Baby AIsBaby AIsCompact, private AI agents built by Vertexpace. Each one runs entirely on your hardware from a tiny file — no cloud, no subscription, no data leaving your device. you edit, recompile, and ship. Each is a knowledge file plus a built-in reasoning engine. ~35 KB native binaryNative binaryA compiled program that runs directly on your hardware. No interpreter, no runtime, no internet connection required. Specialist AIs typically compile to 35–50 KB.. No cloud, no model, no API key. Every answer traces to a fact you wrote.
Genetic programming in Light Code. Give it a goal string. It generates candidate programs using 4 mutation strategies, compiles each to a native binary, runs it, and checks the output — iterating across generations until it finds logic that produces your goal. No model. No guessing. Every answer compiled.
How it works
You define the goal (a string the output must contain)
Factory generates a mutation: DIRECT, QUERY, CONDITIONAL, or CHAIN strategy
Engine shows you the full source code of the mutation
You type AUTHORIZE — nothing compiles without it
Compiled binary runs; loop repeats across generations until goal is hit
Gatekeeper protocol
Hard-coded, not configurable. Every mutation requires your AUTHORIZE before compilation. You are the selector. The AI proposes. You decide.
Deterministic prompt auditor. 12 failure-logic vectors, 8-dimension scorer, LLM-as-Judge cross-model pipeline. Catches sycophancy, hedging, and instruction drift before they cost you. No cloud, no telemetry. Mac ARM64.
70 hand-curated pool servicing answers in one offline HTML file. Chemistry, equipment, troubleshooting, safety. No model. No subscription. Auditable, hedged, yours forever.
35 hand-curated homelab answers in one offline HTML file. ZFS sizing, CGNAT detection, backup that actually restores, container security, the "one weekend" rule. The opposite of a generated wiki.
35 hand-curated answers for backyard beekeepers in one offline HTML file. Bee math, varroa thresholds, queen issues, swarm response, winter prep, and the first-year mistakes documented by extension services.
35 hand-curated first-aid procedures in one offline HTML file. Bleeding control, CPR, choking, anaphylaxis, burns, stroke (FAST), poisoning. Informational only — every lesson defers to medical professionals; call emergency services for any real emergency.
35 hand-curated dog-care answers in one offline HTML file. Feeding, training, vaccinations, body language, separation anxiety, weather safety, when to see the vet — defers to a veterinarian for any medical concern.
1-hour video call. I install Ollama, pick the right model for your RAM, configure Baby AI Lite, and leave you with a working private AI on your own machine.
2–50 compiled specialists running in parallel on a single problem. Each cell applies its domain knowledge independently. Unified output. No LLM. No guessing. Every answer traceable to a fact.
Drop-in Anthropic and OpenAI SDK wrapper. Compresses prompts via the NovaGlyph 64-concept grid before every API call. 13–25% savings on prompts, 50–70% on multi-turn agents. Free. MIT licensed. pip install glyphshrink.
Universal Specialist is the engine — every model below is a custom-compiled variant, built to order in your domain. Order via the button and I'll scope, compile, and deliver. Each is a ~35 KB native binary with auditable FACT-table logic you own outright.
The flagship Developer Kit — the skeleton every specialist on this row is built from, including the $19 browser apps (Pool Pro AI, Homelab Pro) and the $79 native binaries. You write the facts and rules, the kit compiles them. Same answer every time, every answer traceable to a fact you wrote, no model weights to update.
How it works
Edit a plain-text facts file — the "what it knows" — and a short rules file — the "how it decides."
Run one compile command. Output is a ~35 KB native binary or a single offline HTML file — your pick.
Ship it. Runs on a laptop, a Pi, an embedded board, or any browser — no internet, no API key, no telemetry.
Embedded watchdogs for PLCs and air-gappedAir-gapped machinesComputers with no network connection by design. Used in factories, hospitals, and secure facilities to prevent remote access or data leaks. Specialist binaries run on these with zero setup. machines
Contract / lease / lien clause auditors
IoT firmware that needs explainable decisions
Different from…
ChatGPT / Claude — every answer traces to a fact you wrote, not a model guess
LangChain / LlamaIndex — no API keys, no per-call cost, nothing leaves your box
Fine-tuning a model — ~35 KB native binary, not a multi-GB checkpoint
SaaS rules engines — you own the source and the binary outright, no subscription
Tell me what it needs to do and I’ll design, compile, and deliver a one-off specialist binary tuned to your problem. Built on the Universal Specialist engine — same skeleton, your spec, your facts. You don’t touch the toolchain; you get a finished binary plus the source so you can extend it later yourself.
Good fits
You have a domain expert but no compiler skills
The answers must be auditable — no LLM guessing
It needs to run offline, embedded, or air-gapped
You want a fixed-price one-off, not a SaaS subscription
You’re shipping it inside a regulated workflow
You want to own the source outright
What you get
30-min discovery call — I scope the facts file with you.
The cheatsheet that teaches the 64-concept programming language all the specialists above are written in.
PDF
$9
CHEATSHEET
Light Code Cheatsheet
A whole programming language on one page. 64 concepts on an 8×8 grid, plus a second page covering POLYGRID extension rows for file I/O, networking, data, and crypto. 2-page A4 PDF.
Sovereign, air-gapped security infrastructure. No cloud. No subscription. Cryptographic guarantees, not policy promises.
PRO
Enquire
LUMINA_HIVE
Lumina Hive
A digital immune system for machines. Watcher classifies attacks to a 64-concept vocabulary. Defender responds by stakes — autonomous on low, dial-gated on medium, quorum-required on high. Healer rolls back to signed snapshots and repairs faulted code via local LLM. Every decision logged in a tamper-proof audit chain. 3,800 lines of Rust. Ed25519 kinship. No cloud.
Click "Order" on any tile or email robertpage@vertexpaceorganization.com with the model number(s). I confirm price, send an invoice, and ship the signed binary as a download link.
For writing, summarising, brainstorming and document Q&A — about as good as GPT-3.5, sometimes matching GPT-4. Not as strong as frontier cloud models on the hardest reasoning. A fair trade if privacy matters.
To self-install Baby AI Lite: enough to paste one Terminal command. If that sounds like a lot, the $249 Done-for-you tier exists for that reason.
Apple Silicon (M1–M4) and Intel Macs. macOS 13 Ventura or newer. 16 GB RAM minimum, 24 GB+ recommended for the capable model.
Yes. Baby AI Lite only talks to your local AI runtime. No outbound calls anywhere else. Your data never leaves the machine.
If a paid product won't run on your hardware, email me from your receipt. 14-day refund window, no forms.