In the IDE, I install the KILO plugin to use as harness.
I use the Jetbrains Remote IDE feature to code remotely on a cheap linux VM at a cloud provider.
On the remote VM, I have a git repo per project that I open with the remote IDE.
The way I code stuff is to create a separate worktree for each task/prompt and let the agent run at it.
The IDE and KILO plugin lets me watch the progress of each agent and open the chat with each to provide input, instructions, etc. as well as review code changes, perform tests, etc.
I did switch harness at some point because I previously used Cline which was dog slow and unstable.
I do not use OpenRouter or anything like that.
I have dabbled a bit with local models on my MBP M3 but with 24 GB RAM, the models I were able to run were both too slow and poor quality.
My setup is probably hopelessly suboptimal.
I would love to get feedback and hear if there are obvious low-effort improvements.