Hello everyone! I made this app to better monitor changes that agents make in my repositories. It's free and open source - happy to answer any questions!
Hello everyone! Happy to finally present my recent project here on HN :)
The primary problem this software tries to solve is keeping the shape of a project in your head while agents run rampant on it. The more you work with agents the less you understand what's going on. In cotect all changes are easy to locate, you can move quickly through the codebase with WASD and you also get some bonus info like import graphs and basic code analysis.
All code is open source on GitHub and software is completely free. Please let me know if you have questions or feedback!
I think it's difficult to compete in this space because right now to build the "full IDE" you have to build an extremely capable harness (very hard) and a very good IDE (very hard). I plan to continue using Claude Code and built myself a small tool to verify what the agents actually do + move around the codebase quickly but it's a far cry from a full IDE: https://cotect.dev
It's a fast code inspector for agents so that you can easily verify changes before they end up on GitHub and also quickly move around the codebase to keep its shape in your head (to make sure your prompts stay high-signal).
Terminal hunks speed by and I want to review the changes locally before they are on GitHub. Looking through VSC is very slow because it's aimed at writing, not reading, code. In cotect you can move through the repository quickly with WASD and you can approve/comment on hunks that you've viewed and any comments you leave you can copy all at once so you can pipe them down to your agent at once instead of interrupting its work every time you find an issue (or otherwise you forget about that issue - yes I have ADHD).
I was frustrated with not being able to properly review code changes that agents make, looking for a middle ground between VSC and Claude Code. So I built one. Still in early development but its working (mostly) and I hope dogfooding will get me all the way to something great :)
Personally I have multiple instances of Claude on my PC dedicated to different areas (split between my private subscription and my work one) and it is useful to see both at a glance. And I don't have to keep the terminal open to see how much time is left for quota to be reset. But for single-Claude users this is indeed a valid solution!
Doesn't it autocompact only if it is about to run out of context? On these newer models the context is 1M tokens so it's quite difficult to reach. I run compacts manually when I stop in a good place and tokens are over 300k
Seems completely backwards to me. This is like judging Formula 1 just by the raw power of the engine. The rest of the car has just as much engineering, if not more.
Calling autism and ADHD "habits of attention" severely understates what's going on. These are neurodevelopmental conditions with substantial genetic and neurobiological components, not patterns someone can think their way out of. That framing has done real damage to people who delayed treatment because they believed they just needed more discipline or insight. Meditation can be a useful complement for some people, but it's not an alternative to addressing the underlying neurochemistry. By saying so you are hurting real people with real problems.
I'm sure there is plenty of optimization paths left for them if they're a startup. And imho smaller models will keep getting better. And a great business model for people having to buy your chips for each new LLM release :)
This would be killer for exploring simultaneous thinking paths and council-style decision taking. Even with Qwen3-Coder-Next 80B if you could achieve a 10x speed, I'd buy one of those today. Can't wait to see if this is still possible with larger models than 8B.
I really dislike the title and the article. You are basically stating that AI is not a bubble, but everything around AI is a bubble, which people obviously mean when they say AI is a bubble. Same was the case with dot com crash, where people didn't LITERALLY mean .com domains would crash, but everything around the domains (& Internet) would crash.
Does not help that the author is overinvested into the AI space.
Yeah, like asking for trouble. Even when building a basic image upload system in my project I was considering using some AI model to make sure content is at least somewhat moderated.
What's the primary reason you set out to build this? I like the look and feel of the language but I'm on the fence whether the drawbacks of using something different than bash are outweighed by the benefits of elvish.
I've been using it for weeks, but recently it was so gutted that it barely understands TypeScript. I do my own coding again because sifting through all the possible bugs is more work than just writing it yourself. A true shame, because before it got "faster" it was capable of creating entire apps.