I built Cortex66, so read this as a founder writing about their own project.
The problem I kept having
For months my way of working with AI was the same loop. Ask one model for a draft. Paste it into a second one to check it. Take the result to a third for another pass. Every step was a copy and paste between browser tabs, and I was the messenger.
The second problem was money. Each of those tools has its own bill or its own usage meter. I never knew what one piece of work had cost me until much later, and by then I could not connect the number to the task.
What I wanted instead
Two things.
First, give one goal and have it split into steps, with each step sent to the AI that fits it, using the accounts and keys I already have.
Second, see the price before anything paid runs. Not a dashboard at the end of the month. A number next to the Run button, and a cap I set myself.
What I learned building it
A cost you see first is a decision. A cost you see later is a surprise. The hard part was not calling the models. It was producing an estimate early enough that a person can still say no. An estimate is only an estimate, so the cap matters more than the number: the run may cost less, but it may not cost more.
A plan you can read is easier to fix than one you cannot. When the split into steps is shown as a short list, you can see the wrong step before it runs. When it is hidden, you only see the wrong result.
A second model checking the first one catches real mistakes. One writes, one tests, one checks. It is slower than a single answer and it is not magic, but the checking step is where I got the most value.
Your own keys should stay yours. If you already pay a provider, the tool should use that and not charge you again for the same call.
What is still rough
I would rather say this plainly. Team runs with several real AI apps at once are new. The docs are thin. Estimates are estimates. It is a solo project, about a month old, and AI agents wrote much of the code under my review.
If you want to look
The project is at https://cortex66.com and there is a CLI on npm (npx cortex66). You can start free with your own keys.
What I would most like to hear: when you use more than one AI on the same task, how do you keep track of what it costs you?
Top comments (1)
the cap is the part i'd actually trust, estimates drift but a hard ceiling i can refuse before anything runs is what makes handing the run over feel ok. same reason we built searchspot to catch the expensive trip mistakes before you pay instead of after, link's in my bio if you want to try it