I’ve been experimenting with Vibe Kanban for the past few weeks, and honestly it feels like stepping into a sci‑fi cockpit where every button press automatically runs a whole pipeline of code. The tool sits at the intersection of three things I love: kanban boards, Git version control, and AI agents that can actually write code.
Why Vibe Kanban Now?
Traditional kanban apps are fantastic for visualising work items, but they stop short at “move this card”. Behind each move you still have to:
- Create a branch (`git checkout -b feature/foo`)
- Edit files in your editor – Run tests (`npm test`)
- Push the branch (`git push origin …`)
- Open a pull request on GitHub/GitLab
All of those steps are manual, error‑prone, and—let’s be honest—boring. Vibe Kanban automates every one of them.
How It Works (In Plain English)
**Install** – `npx @vibe/kanban` gets you a UI running in seconds. No global dependencies, no Docker containers unless you want them.
**Connect a Git repo** – either clone an existing one or initialise a fresh folder with `git init`. Vibe Kanban insists on a repository; cards can’t exist without it.
**Create a card** – give it a title, pick a backlog column, and hit *Start*
**Move the card to “In Progress”** – behind the scenes Vibe spawns a new branch, checks out that branch, and fires off an AI model you’ve configured (Claude, Gemini, Codex… or even a local LM Studio instance). The AI does the actual coding, runs tests, writes documentation—whatever your card description asks for.
**When you’re ready**, click **Merge**. Vibe automatically merges the branch into `main`, updates the board to “Done”, and you can move on.
All of this happens with a single mouse movement. No terminal, no cryptic Git flags—just a clean visual workflow that feels almost magical.
Setting Up Your Own AI Profile
What really sets Vibe Kanban apart is its **profile system**. Profiles live in `profiles.json` and let you map any OpenAI‑compatible endpoint to a UI entry. Want to use the local LLM you’re running on LM Studio? Add something like:
{
"label": "lmstudio-local",
"OPENAI": {
"command": {
"base": "echo",
"params": []
},
"env": {
"OPENAI_API_BASE": "http://localhost:1234/v1",
"OPENAI_API_KEY": "not-needed"
}
},
"mcp_config_path": null,
"variants": []
}Save the file, restart Vibe Kanban, and you’ll see “lmstudio‑local” appear in the dropdown. From that point on, any card can be assigned to your local model with a single click.
Who Should Give It a Try?
Branch‑hopping developers** – if you constantly juggle feature branches across services, Vibe Kanban’s automatic branch handling will feel like a lifesaver. **AI enthusiasts** – finally you can delegate real coding tasks to Claude or Gemini without leaving your board. **Teams with mixed Git proficiency** – anyone can move cards safely; the system guarantees that every move results in a proper Git operation.
My Takeaway After a Week of Use
I was skeptical at first—how could moving a sticky‑note‑like card actually generate real code? The answer is simple: **the card *is* the command**. When I moved “Add user authentication” from *Backlog* to *In Progress*, Vibe created a branch, called Codex (via my local profile), wrote the entire login service, ran `npm test`, and opened a merge request—all without me typing a single character.
The experience reminded me of playing with a remote‑controlled robot arm: you guide it with simple gestures, but the heavy lifting happens behind the scenes. And unlike most “AI‑assisted” IDE extensions, Vibe Kanban works **outside** any specific editor. You can stay in your favourite text editor, let the board do the orchestration, and only open a file when you need to review or tweak something.
What’s Next?
The project is still young, but the roadmap promises tighter integrations with CI pipelines, richer state machines for cards, and even support for non‑code tasks like drafting blog posts or generating meeting minutes. If you’re curious about how a kanban board can become a full‑blown AI command centre, give Vibe Kanban a spin. Install it locally, add your favourite model, and watch Git dance to the rhythm of your visual workflow.
If you enjoyed this walkthrough, feel free to share your own experiments in the comments below. I’m especially interested in hearing how different AI models behave when they’re orchestrated through Vibe Kanban’s card‑based interface.*
