Open Code: The Minimalist Terminal Agent I Actually Use Daily

I've been writing code for three or four years now, and most coding agents feel like bloatware at this point. Cursor is built on Electron and the resource usage is insane. Claude CLI is decent but still feels like a work in progress, plus you need a Claude subscription. Other tools like Roo Code need VS Code open just to function.

I used to run Arch Linux with Hyprland and recently moved to Omarchy as it has good community now. I needed something that fits that workflow. GitHub Copilot and Gemini weren't good enough. Then I found Open Code.

Why I switched to Open Code

Open Code has come a long way in the last two years. It now does everything Claude CLI does, Skills and MCP servers included. When I want to add a feature, fix a bug, or tackle a GitHub issue, I just:

Open a tmux session.
Launch Open Code.
Minimize the terminal and do something else while it works.

No full IDE sitting open chewing RAM. It just runs in the background. It's also open source. Any skills that work with Claude work here, and you can write your own custom agents.

Pricing and modes

I use Open Code Go at $10 a month. The limits are fine for me. I tinker with open-source stuff and hunt for bugs on weekends, and I haven't hit a wall yet.

Agentic Modes are my favorite part. Build and Plan modes come out of the box, but I made a custom Security Engineer mode that checks for security flaws in the code. You can create any mode you want. The Security one is what I use daily.

Performance and integration

Since it runs in the terminal, all my themes match out of the box. Tool calling got a lot better with models like Claude 3.5 Sonnet and GPT-4o.

Privacy

This matters most to me. Open Code talks directly to the AI providers. No middleman server scraping your data like some other CLIs do.

MCP support

Open Code supports Model Context Protocol. Say you want to visualize something. Add the Excalidraw MCP, give the agent a task, and it draws or presents data for you right there in the terminal.

I still think AI has long way to go. For now, it takes of lot of repetetive tasks and simple issues that do not nee much oversight.

Until next time PEACE

Surya Chilukuri