I've spent a lot of time dabbling with agentic ADEs and IDEs since the days of the early VSCode forks. I've switched IDEs more times than I can count. Cursor. Copilot. Codex. PearAI. Orca. Claude Code. OpenCode. Superset. Agent Orchestrator. Windsurf, and then Devin. OpenChamber. Kilo. IntelliJ and VSCode plugins.

This September, I felt sort of stranded. I kept switching tools, losing all my chats, losing time, and frankly, losing love for the very thing that has been getting me out of bed every morning for the past decade. Programming. Building. Tinkering. I've spent more time fighting my tools, being forced back into terminals, being rug-pulled by labs, or being forced into the cloud, when really, I just want to work on my machine.

It was then that I started to jot down my frustrations in a blog post. Not that anyone would read it, especially these days, but my frustrations, I felt, needed an outlet. As I was writing the post, I sort of realised: for the longest time, I'd been observing and obsessing over products like Linear. Raycast. Arc. Their level of detail. Their thoughtful approach to UX. Their grassroots, scratch-a-boring-itch story.

So I decided to scratch my itch, and build an IDE that would stay out of your way. It's called Althar. Delightful to use, calm, and open, rethinking the old paradigms that are holding most of the industry captive. There's no sidebar. There's no built-in text editor. There are no chats (in a traditional sense).

Agents forgetting my project between tasks always annoyed me. So did my overarching goals and milestones getting lost between chats, and agents having no way to coordinate work with one another. In most tools I've tried, they work in silos.

So, in Althar, you organise your work in projects. A project can be a single repository, or many. On top of it sits your project manager, called the coordinator. You talk to her like you would talk to any PM on a project. You could say something like, "Help me make my website responsive and accessible. Also, help me measure the performance with Lighthouse." She will then split the work into three separate tasks, brief individual task agents called leads, assign each task a reviewer agent (we love a second opinion, but it's optional), and send them on their way.

Finished work comes back to you with screenshots, test evidence, and the reviewer's findings settled.

You see everything that's happening on a kanban-like board (though it is NOT kanban), and you're only pulled in when you're actually needed. No need to anxiously watch over terminals and in-flight agent thinking trails.

You can use all of your existing subscriptions. Claude. Codex. OpenCode and so on. It connects to your issue trackers and git hosts, wherever they may be.

Work happens locally. Next week, I hope to add support for remote runners, and I'm also working on a mobile companion.

I hope you'll enjoy it. There is so much to say, so many little details. But it's also early days, so please, if you do find issues, let me know :) Or open a pull request. The open source community has given me so much over the years. It is the best thing about software engineering, and something unique to our profession. I'm excited that I get to give back.