Breaking Free from AI Lock-In: Testing Droid by Factory AI vs Claude Code and Copilot CLI

Breaking Free from AI Lock-In: Testing Droid by Factory AI vs Claude Code and Copilot CLI

Claude Code made waves when it first released. It was one of the first tools designed to let you build an entire app without ever opening your IDE. Whether that’s advisable or not is up for debate, but I’ve always found tools like Claude Code useful, especially because I prefer a terminal-first workflow.

Apparently, I wasn’t the only one who liked what Claude Code had to offer. It gained enough traction that Microsoft recently dropped its own competitor: Copilot CLI.

The Problem with AI CLI Tools: Vendor and Model Lock-In

While both tools are solid and deliver on their promises, they share the same major downside: vendor and model lock-in.

If you’re using Claude Code, it’s no surprise that (a) it’s a paid tool, and (b) you can only use Claude models with it. The same goes for Copilot CLI. You can’t choose your model, and it’s bundled into an expensive Enterprise plan.

That’s frustrating, right? Limiting, even.

Enter Droid by Factory AI

This is where Droid by Factory AI comes in. A third option that changes the game. Yes, it has a paid plan, but here’s the kicker: you can bring your own keys. 

That means you can use any provider you want. Claude? Sure. OpenAI? Absolutely. Even local models.

That last part is what really got me excited. So excited, in fact, that I spent an entire Sunday experimenting with different local models. (That’s a story for another blog.)

Does It Actually Work?

Short answer: yes, and it’s awesome.

I had a wild idea to build a command shell, but instead of Bash or Zsh, it’s written in Golang. This project has been equal parts exciting and challenging, and I’ve leaned heavily on Droid to bring it to life. (By the way, version 0.1.0 of Gosh is up on GitHub.)

Since I’m currently using the free 20-million-token, 30-day trial, I went with Factory’s Droid Core GLM 4.6 model. It has a low token multiplier but still performs like a powerhouse.

All in all, it’s a flexible, high-performing tool that fills a unique niche: supporting models from multiple sources without tying you down.

Your Turn

Have you tried any of these AI CLI tools: Claude Code, Copilot CLI, or Droid by Factory AI?

What’s your experience been so far?