Maisie Eleanor

Hey PH We're Baptiste & Samuel, and next week we're launching Capacity Desktop a free Mac app that turns plain English into real apps, built and stored on your machine.

The bet everyone tells us is wrong: move away from credits system. You plug in your own Claude / GPT / Open Router key and pay your provider at cost. Our take is that you feel the credits system is unfair, you never know what a failed fix loop just cost you, and you feel the tool profits from your retries.

For me, I feel like the hardest part isn't choosing between tools like React, Firebase, Railway, etc, it's really deciding which parts of the stack I should commit most of my time and tokens early and which ones may become painful to replace later.

Do you stick with a familiar stack and avoid overthinking it, design around expected scale from day one, let your coding agent influence the architecture, or start with the cheapest setup and migrate later? Also, curious which early technical choice has saved you the most time, and which one you wish you probably shouldn't have overthinked on.

When we started NeuralAgent back in April 2025 it was one of the first agents in the world that uses a user's personal computer, not a computer on the cloud, thats what made it to spread to 180 countries and tens of thousands of users in a short amount of time.
Today, I am excited to announce that we are experimenting with something new, something called NeuralOS, it's the next evolution of the Neural experience.
With NeuralOS, intelligence stops being an app you open and close and it becomes the OS itself, it becomes the computer itself, it's always on and always ready. Imagine for example you are now on the gmail interface, and you say 'Hey Neural, send a reply to John here', and it just understands, it just knows and in milliseconds types the reply, asks for your approval and then sends it. Or if you are on an interface and there was an error and you say "Hey Neural, fix this" and it just knows! It just works across any app, any screen, any interface, you call it via a keyboard shortcut or voice and ask it to do something, and it does it in milliseconds.
Now that we have the Neural fast model, I feel now is the time for this experiment.
Stay tuned for NeuralOS! It will be an OS layer that's always on that can run on Windows and macOS and soon Linux as well!
As always, we would love to hear your feedback!
Check out the NeuralOS demo video here: