Lê Quang Phú
Back to blog
•
2 min read
Terminal + LLM: from Warp to Ghostty

I am using aichat and a local LLM inside Ghostty to replace Warp. Why?

  • Warp’s app icon is ugly, and Ghostty’s app icon is beautiful.
  • I challenge myself to work with a local LLM instead of a free version of a paid LLM product. I have to connect the local LLM to Ghostty via aichat.

At the time of writing, the local LLM is qwen3-4b-2507 served by LM Studio. The model is the best among models of its size for me. I used Grok (x.ai) to install Ghostty, aichat (rustup + cargo), and LM Studio. I recommend model configs as below.

Model configs

6 bits quantization is recommended by an Apple’s staff working on their AI.