I built an app I use every day. I charged $20 for it, once, yours forever. Fair price. Zero users.
Not "low traction." Zero. I couldn't give it away — I handed out promo codes for free copies just to get my first five users, and even that didn't work. People DM'd me asking for access, saw the price, and bounced. "Herdr and T3 Code are good enough for me." Gone.
So AYGENT Core is now free and open source under the Apache license. This is the story of why, and what I'm keeping behind the $7.99/mo Pro tier — and why I think that's the only honest way to sell software in the agentic era.
It started with OpenClaw driving me insane
I didn't set out to build a harness. I was an OpenClaw user, and two things about it wore me down.
First, maintenance. Every time a new model dropped, I wanted to try it immediately — and switching models meant going in and fixing config files by hand. On a VPS, on Linux, the thing that was supposed to hot reload. I'd ask the bot to switch the model, it would half-work, and then I'd be elbow-deep in configs. For a tool whose whole point is that agents do the work, I was doing a lot of work.
Second, money. OpenClaw was terrible at managing cache reads on Anthropic models, so I was paying way more per token than I needed to. When I finally built my own harness, my token costs dropped 80%. Eighty percent — for the same work.
And third, the one that bothered me most: I lived in Telegram. Just a chat window, all day, with this weird disconnect between me and what the agent was doing. I couldn't see what it was working on. No Git diffs. No code. Getting a file out of it was a chore. I kept thinking: I'm talking about the work instead of looking at the work. That could be so much better.
So I built the thing I wanted: agents on my machine, memory in files I own, every action visible and undoable. A native Mac app. $20, once. I figured "dramatically easier than OpenClaw" would sell itself.
The thesis changed under me
Here's what I believe now: we're entering an age where anything can be built by anyone — usually tailored to their own workflow, usually cheaper than buying it. If there's a subset of Photoshop features you use daily, you can spend a day cloning exactly those, shaped exactly how you want, and skip the subscription. I think most people using agents are going to build their own software. That's not a prediction anymore; I'm watching it happen.
I saw a project — built with AI — attempting to clone the entire Adobe suite. Free and open source. To be clear: I didn't make it, I'm not planning to make anything like it, and I honestly don't know what the legal exposure on something like that looks like. But it exists, and it's probably unstoppable, and that alone changes the math for everyone.
Think about what's actually empowering here. Twenty years of proprietary Adobe development, and now there's an open codebase anyone can read, review, add to, fix — things you fundamentally cannot do with Adobe. If I wanted to, I could download a Premiere clone tomorrow, keep what I like, build what I need on top of it, and cancel my subscription. That's agency over my own workflow, and that agency is the market shift. It's changing the face of software right now.
These clones are the canary in the coal mine. We're almost at the point where you can't charge outrageous prices even for genuinely great software, because someone can read your feature list and copy it. Give AI a couple more years of getting better and cheaper, and we're not paying for apps at all — it just builds them for us.
And that kills the moat for something like an agent harness. Anyone who understands agents can build one for not that many tokens. Everyone I've talked to who would use AYGENT would honestly rather build their own. $20 was palatable! It didn't matter. The price wasn't the objection — ownership was.
There's a second shift, and it's about trust. When software has deep access to your machine and your personal life, and AI agents act on your behalf, people want to inspect the code. Especially on the frontier, where everyone is technical: everything needs to be observable. You read the code, you decide if you trust it. A closed $20 binary from a guy on the internet fails that test before it starts. My harness got better the same way — by reading open source code and asking "how do I make this better than theirs?" Everyone's going to do that. If someone finds your feature list, they'll just ask their bot to build it. Unless you're shipping something truly proprietary, software doesn't have a moat anymore.
So the logic inverted: if it's free, there's no token cost and no time cost to trying it. You download it instead of building it. And if you've already gotten real value from the founder's free work, and they offer a couple of well-thought-out extras for cheap that make starting easy — maybe you pay for that. That's the experiment I'm running.
What's free, and what's Pro
Core — chat, agents, tools, scheduler, memory, the whole engine — is Apache-2.0. Contribute to it, fork it, learn from it, security-test it. I genuinely want contributors and credibility more than I want $20.
Pro exists for two reasons, and both are things I couldn't open source without killing the business reason to exist:
1. Inference with my taste built in. I have strong opinions about which model does which job, and Pro lets you rent mine instead of managing five API accounts. Three tiers, matching how the labs themselves segment it:
- Expert runs Opus 5.5 right now — the best all-around model for hard coding work, reliable at running a loop until the job's done. My actual workflow: I run everything on the workhorse, and when I'm chasing my own tail on something truly hard, I pay ~$7 to let Opus fix that one problem, then go back.
- Workhorse is Muse Spark 1.3. Most of my work happens here. It's in that tier because it's where I live.
- Intern is also Muse Spark 1.3, but the contributor tier — your usage can train Meta's models, which makes it extremely cheap. I could've used DeepSeek; it just doesn't do real tasks as well. Contributor Spark at rock-bottom pricing is the best cheap intelligence you can actually work with.
And because the app sends tiers, not model IDs, I can swap the model behind any tier the day something better drops. No app update, no config surgery — the exact pain OpenClaw put me through.
2. The agentic video editor. Core ships a Premiere Pro MCP, and it's decent — but Premiere measures timecode in proprietary ticks that don't behave the same between video and audio, so I kept chasing missed frames and mis-timed audio across two apps, fixing every little nuance by hand. Death by a thousand frictions.
The Pro editor is a different take: the project is a simple JSON file, the edit happens through code libraries inside the app, and the whole thing was built by the agent that edits with it — so it genuinely understands what it can and can't do. A decade of editing went into making it the talking-head-plus-graphics-plus-sound-design editor I always wanted. Working with it feels like hiring an editor: you ask for a cut, you give feedback on the timeline with tools I built for exactly that, it revises. Reviewers never open Premiere — but your agent has the timeline open, with full project context, in its own codebase.
The trough
I'll be honest about where this sits: I'm in the trough of disillusionment. Zero users. I built something I truly like using — built exactly to my taste, does everything I want, shockingly reliable — and almost nobody has looked at it. Maybe I built the perfect custom tool for me. Maybe people don't want agents. Maybe I need to go harder on marketing. Truth yet to be seen.
But here's my thesis on transparency: saying "zero users" out loud makes me more personable, not less. Maybe you read this, maybe you like my stuff, maybe you figure — it's free, why not try it? That's the long game: make something great, give it away, build a reputation as a builder, earn sponsors or donations or client work or funding or wherever it goes next. Free gives it a life. Paid gave it a gravestone.
If any of that resonates: the repo is here. Read the code first — that's the point. If you like what you see, download Core free, and star it so I have some credibility to show for all of this.