Private
Conversations, files and memory stay on private hardware. Your data is not being saved and harvested by the tech giants.
Lane 03 · AI
Built here first.
Built for you next.
This site was built with AI. So are the agents, apps and creative projects RoadCodez takes on next — business automation, apps and tools, photo and video work — all with a human signing off before anything goes out the door.
Make no mistake, AI is an incredibly powerful tool, but it is just that… a tool. Without a human assisting through competent design, implementation, review, and quality control phases, an AI can make mistakes. Imagine AI as a tool in your toolbox. Without a human to turn the wrench, swing the hammer, or run the saw… those tools are useless.
Why local-first
Your data. Your rules.
Predictable cost
Open models run locally; a cloud model only gets used when a task actually earns it.
No lock-in
Your data lives in your own database, so you can swap models or vendors without starting over.
Auditable
Every action an agent takes is logged, so you can see what it did, when, and why.
How it works
A manager, specialists and a gate
The AI model never has the last word. A deterministic control layer decides what’s allowed, what needs a human, and what is simply denied.
Step 1YouAsk by voice or text, from your desk or your cab.
Step 2Local manager modelUnderstands the request, plans the work and decides who should do it.
Step 3 · The control layerPermissions & approvals
Plain, predictable rules — not an AI guess — check every action before it happens.
Allow — safe, sandboxed workAsk a human — pushes, spending, messagesDeny — anything unknown
Step 4Specialist modelsCoding, reasoning and research, each handled by the model best suited to it.
Step 4Sandboxed toolsFiles, code and web access inside fenced-off folders, never the whole machine.
Always onLong-term memoryA local database that remembers tasks, decisions and context across sessions.
Where things stand
Building in the open
Several things are moving at once — agents, apps and creative tools, including this website. Specific case studies go up here once a couple are finished, not before.
Business & CRM automation
In progressWiring up a CRM so a one-person business can find contacts and automate the busywork it doesn’t have time for — built alongside a developer friend who owns the CRM itself.
A driver’s own app
In progressAn app for my own truck: tracking the business side of driving, and talking to it — by voice — while I’m on the road.
Community app
In progressA community-focused app, still early. More on this once there’s something worth showing.
This website
OperatingBuilt, written and maintained with AI — branches, review and all — from the ground up.
What I take on
Agents, apps, and anything creative
Business-process agents
An agent that runs a repeatable process — intake, routing, follow-up, whatever it is — and hands you the result instead of the busywork.
Apps & tools
End-to-end app development: a community app, a tool built around one specific job, something you can talk to instead of type into.
CRM & business automation
Making the systems a small business already has actually work for it — finding contacts, automating the repetitive parts, so the owner can focus on the job.
Photo, video & design
Image editing and generation, video work, and design — AI doing the production, a person picking the final cut.
A small example
From a bare skeleton to this
A real example of what AI image generation can do, not a staged demo.
I uploaded a plain skeleton on a black background and asked ChatGPT to dress it in everything it knows about me. That’s the whole process — no manual editing on top. Combining commercial AI with a goal of privacy-based data storage is the key.
It’s a small, fun case, but the same idea scales up to real work: product shots, social content, or cleaning up a photo that almost worked.
Ground rules
How I build with AI
- Local firstRun on private hardware by default; reach for the cloud only when it’s worth it.
- Human in the loopPublishing, spending money, sending messages and changing networks always need a person’s approval.
- Fail closedIf an action isn’t explicitly allowed, it’s denied.
- Test before trustGenerated code is checked and tested before it’s accepted — the model’s confidence isn’t proof.
- Audit everythingEvery action leaves a record that can be read back later.
Let’s talk
Got something to automate, build, or make look good?
Tell Ogre what you’d like off your plate, built, or made to look right.