An app on your computer for building software with two AIs.
ProxiLoop puts a builder AI and a reviewer AI to work on your project, best from two different companies. You approve the plan, and you decide when the work is finished.
Launching November 1
Windows and macOS versions are in development and will be released as soon as testing and validation are complete.
You don't need to become a programmer.
You need to learn how to direct the work.
Your tickets and the step each one is at, what the reviewer has found, and the two AIs' own windows side by side, where you talk to them directly.
Every decision is saved with the reason it was made, and every lesson from what went wrong, so each new chat starts out knowing them.
You look at the plan before anything is built.
ProxiLoop opens it where it is and adds its own filing beside your work, without overwriting anything. From then on, the builder works from plans you say yes to.
You can switch either review off. Lock and Ship stay yours either way.
Split into the ones caught in plans, before any code was written, and the ones caught in finished work.
Both AIs explain in plain words, then list your next steps, numbered, in order.
ProxiLoop is built for the builder and the reviewer to be different AIs, made by different companies. Only have one AI? You can start with it. It plans, builds and reviews its own work, and it still catches real problems.
ProxiLoop works best when you add a second AI from a different company, because it can catch the mistakes the first one tends to miss.
An AI checking its own work tends to agree with itself. The reviewer reads the plan and the work fresh, with no stake in either being right.
Two AIs from the same maker tend to share the same mistakes. A reviewer from a different company looks at the work differently.
The plan is reviewed before any work starts, when a wrong idea is still only words. The finished work is reviewed again.
An AI reporting on its own work tends to round up. The reviewer checks those claims against the actual work and the test results.
I once used up my reviewer AI's usage before the month was over, and I needed to keep building. For five days, I used the same AI as both my builder and my reviewer. It still caught real problems, even though it was checking its own work.
When my usage came back, the second AI's very first review caught a kind of mistake the same-AI reviews had not been catching.
You use your own accounts with each AI. Any of them can be the builder or the reviewer. Two from different companies work best.
Not sure which to put where? Pick either way. You can swap the builder and the reviewer between tickets.
| AI | Made by | Status |
|---|---|---|
| Claude CodeAnthropic | Anthropic | ✓ Works today |
| Grok BuildxAI | xAI | ✓ Works today |
| OpenAI Codex CLIOpenAI | OpenAI | ○ Planned |
| GitHub Copilot CLIGitHub | GitHub | ○ Planned |
| OpenCodeAnomalyCan use many models, including open-source ones through services such as OpenRouter. | Anomaly | ✓ Works today |
| Antigravity CLIGoogle | ◇ Under evaluation |
Planned AIs have no date yet. "Under evaluation" means we are checking how it can be used inside other apps before we plan it.
Writes a plan for the work you asked for.
Points out problems before anything is built.
Approve the plan.
Does the work in the locked plan.
Checks the finished work against the plan.
Decide when the work is done.
Plan review and Build review are on by default. You can switch either one off, for one ticket or for the whole project, with one press on the page. Lock and Ship stay yours either way.
The builder is an AI such as Claude Code. The reviewer is an AI from a different company, such as Grok Build. Lock and Ship are yours. The AIs are set up to wait for you.
Every plan, review and decision is kept in your project, and ProxiLoop shows you the step each piece of work is at.
Pick it up tomorrow or next month. Your project keeps its own notes, so the AI catches up on where you left off, what was decided and why. No re-explaining.
Every decision is saved with the reason it was made, so you're never left wondering why something was built a certain way.
When something goes wrong, the lesson is written down, so the same problem doesn't catch you twice.
Nothing is approved or finished until you say so.
You direct the work in plain words, like a project manager, not a programmer.
Plans, reviews and decisions are saved as files in your own project folder, not in someone else's cloud.
119 of them were in plans, caught before anything was built.
That is why the plan is reviewed first: a mistake caught in a plan takes minutes to fix. Caught after the building, it means redoing the work.
The setup steps would have left someone on a new computer going round in a loop between two steps, forever. The builder's own check would have passed.
A warning said one more file was hidden when there were six. The builder's own test said everything was fine.
A screen still promised a save would leave the project clean, after a change that meant it wouldn't.
None of these would have shown up as an error. Each one would have reached a real person.
I'm Saul, a Navy veteran who spent more than ten years managing fiber and telecom builds on projects for Verizon, Google Fiber, AT&T, and Comcast. As a project manager, my job was never to lay the fiber. It was to keep all the moving parts of a project working toward the same goal.
My nephew got me looking into AI, and I was amazed at what it could do. I started experimenting with building apps, and through a lot of trial and error, I found myself running them the way I'd run any project. I treated the AIs like a skilled, tireless crew.
ProxiLoop is that process, and you choose your crew. One AI builds, another checks the work, and you decide what ships.
A computer running Linux (Windows and macOS are coming soon), and your own accounts with the AIs. Claude Code, Grok Build and OpenCode work today, and more are planned: see Which AIs work with ProxiLoop.
Yes. One AI can do both jobs, and it still catches real problems. ProxiLoop works best with two from different companies, because the second one can catch what the first one misses.
ProxiLoop itself sends none of your work anywhere. The AI programs send what you ask of them to their own companies, as they do whenever you use them.
No. ProxiLoop is made by AetherMindForge. None of the AI companies named here made it, sponsors it or endorses it.
I'll email you when ProxiLoop launches.
Pricing: coming soon.
Questions? Email me, I read every message: hello@proxiloop.com
