I build AstraCode. What I wanted was an editor that finishes. Most AI tools write plausible code and hand it back for you to find out whether it works. This one plans the change across the repository, writes it, runs your tests, and only then shows you a diff so the first thing you see is a change that already passed. The task we benchmark against is deliberately small: one localized defect in a TypeScript repo. It takes 14 requests and reads 408,878 tokens of context to get right. That number is where the rest of the product came from and it turns out several frontier models charge a higher rate on the ENTIRE request once the prompt passes a token line, and one ordinary run clears it. So the editor prices each step at the rate that actually applied, and shows you the number as it goes. Honest gaps: macOS and Linux, no Windows yet. It's new, so the ecosystem around it is small. Where other tools are the better buy I've written that down instead of pretending otherwise the comparison pages concede on price twice. If you want to poke at something without installing anything, the cost calculator is free and needs no account: astracode.io/cost Happy to answer anything, including the awkward questions.

AstraCode

An AI code editor that checks its own work

An AI code editor that checks its own work