Castforge

I started Castforge in March with a simple pitch: you already pay for Claude Pro and ChatGPT Plus, so use those subscriptions instead of paying per token through five dashboards. Bring what you have and start coding.

It was a good pitch. It was also the wrong foundation, and I did not find that out from a customer. I found it out from a provider pricing page.

The terms around programmatic use shifted, and that whole angle stopped being something I could honestly build a business on.

That is a strange thing to absorb two months in. The product still worked. The sentence I had been using to explain it did not.

What I did was stop selling the billing and start selling the thing I had accidentally spent all my time on.

Because when I looked at where the months had actually gone, almost none of it was auth. It was making a team of agents tell the truth about what they had done:

A board that does not claim a card is finished when nothing was recorded.

A check that says it could not run rather than quietly passing.

Opening a project not being an accidental decision to spend money.

Nobody was ever going to stay for the billing. They stay because the thing does not lie to them.

I had been building the real product for months while describing a different one.

Two things I would tell anyone in the same spot:

Write your premise down somewhere you will re-read. Mine lived in my head, so when it died I had nothing to diff against, and it took weeks to notice.

Watch where your hours actually go, not where your pitch says they go. The gap between those two was the answer, and it was sitting there in my own commit history the whole time.

Castforge is in public beta now. Windows. Free while it is in early access.

Solo, six months in.

castforge.ai