Manual IQ
Your documents. Your machine. No cloud. No hallucinations
2 followers
Your documents. Your machine. No cloud. No hallucinations
2 followers
Ask plain-English questions about your technical documents. Get cited answers with page numbers and section references. 100% local via Ollama or cloud via Claude — your choice. Diagrams surface inline from the source PDF. Built for pilots, nursing students, law firms — anyone whose documents can't go to ChatGPT. $59 one-time. Code PH29 saves $30.
On a Wednesday morning in April 2026, I sat down and decided to build something.
I'm a professional pilot and flight instructor. For years I've watched colleagues dig through 600-page PDFs mid-brief, hunting for a number buried in a table on page 312. I always thought: there has to be a better way.
The better way exists. It's called RAG — Retrieval-Augmented Generation. AI reads your document, indexes it, and answers your questions with citations. The tech isn't new. The application to proprietary, licensed documents — running 100% locally so nothing ever leaves your machine — that part is.
I built the first version in a day. Tested it on a Praetor 600 QRH. Asked it for the maximum oil consumption limit. It came back in seconds: 0.04 quarts per hour — QRH page 47, Section 2.3.
Thirty-six hours after I started, we had our first sale.
That's Customer #1. That's CAVOK Designs.
We're building local-first AI for professionals who work with documents they can't share with the cloud. If that's you — welcome. Follow along. There's a lot more coming.
— John Tubbert, Founder, CAVOK Designs
One rule I built every agent around:
"The document teaches the system first. The user teaches the system second. The developer teaches the system never."
Pricing: $149 one-time. Use code PRODUCTHUNT at checkout for 50% off — good today only. No subscription, no cloud fees, no per-query charges after purchase.
Here's ManualIQ answering "what is the crosswind limitation" against a Praetor 600 AFM — running completely local on a 5-year-old M1 MacBook 8GB using gemma4:e2b, a 2 billion parameter model.
No API key. No network calls. Four precise page citations.
The model doesn't need to be powerful because ManualIQ finds the right pages first. The inference is just formatting what's already there.
v3.1 is live. Mac Apple Silicon stable. Windows version in testing. Early adopter price with code PH29 at cavokdesigns.com