Back to Blog
Engineering

Karna: Your Independent Coding Buddy

April 12, 2026
4 min read
AgentKanbanRustNextjs
Karna: Your Independent Coding Buddy

I have a problem. I get ideas constantly. In the shower, on a walk, mid-conversation with friends. Some app concept, a tool that should exist, a side project that would genuinely be fun to build. My notes app is overflowing with them.

And here's the thing: most of them aren't even hard. They're not moonshot research projects. They're a weekend of focused coding away from being real. But I already write code all day at my job. When I close my laptop in the evening, the last thing I want to do is open another terminal and grind through boilerplate for a hobby project. I want to go outside. See my friends. Cook dinner. Read a book. Be a person.

So the ideas just pile up. Not because I lack the skill. Not because the projects are too ambitious. But because even a simple build requires me to sit there, set up the repo, wire the pieces together, debug the silly stuff. It's not hard work. It's just time. Time I'd rather spend living.

I built Karna because I was tired of choosing between my ideas and my life.

What It Actually Does

Karna is a self-hosted autonomous coding agent. You give it tasks on a kanban board, and it does the rest. It plans an approach (which you review whenever you get a minute), implements the code, and opens a PR on GitHub. If something needs changing, you leave a comment and it iterates. The ideas actually get built now. And I get to keep my evenings.

This is for the person who gets a new idea every ten minutes but also genuinely enjoys not being at a computer. You love building things. You just don't want building things to consume every waking hour. You already code eight hours a day for work. Your side projects deserve to exist, but they shouldn't cost you your weekends and your sanity.

What Makes It Click

You describe it, the agent builds it. Open the board, type what you want, pick a priority. The agent handles branching, implementation, and PR creation. You review when you feel like it. Maybe that's over morning coffee. Maybe that's three days later. It doesn't matter.

Plans before code. The agent doesn't just run off and build something wrong. It writes a plan first. You skim it, give a thumbs up or redirect, and only then does it start writing code. Five minutes of review saves you from "that's not what I meant" rewrites.

Pick your AI. It shells out to Claude Code or OpenAI Codex under the hood. You can choose the backend and model per task. When the next great coding AI drops, adding support is a small adapter away.

Multi-repo, because real projects aren't one folder. Most ideas need at least a frontend and a backend. Karna automatically profiles your repos, understands their structure, and splits cross-repo work into subtasks that each go through their own plan-and-review cycle.

Scheduled scans. Set up cron jobs that hunt for bugs or flag improvements in code the agent already wrote. It maintains its own output. You just check in when something interesting surfaces.

It can improve itself. Point Karna at its own repo and it will fix its own bugs, add new skills, and update its own instructions. Recursive delegation, all the way down.

The Nerdy Details

The agent is written in Rust for fast, lightweight polling. The frontend is Next.js 15 with a clean kanban interface. Everything runs in Docker Compose on your machine. Your code and API keys never leave your network. There's even a browser-based VS Code so you can watch the agent work in real-time if curiosity gets the best of you. But the whole point is that you don't have to watch.

The Name

Karna, from the Mahabharata. A warrior who was self-made, fiercely independent, and relentlessly capable. It felt right for something that fights on your behalf while you're out living your life.

If you're someone with too many ideas and not enough free time, give it a spin. It's fully open source. Head over to GitHub now and start making!