What's your environment?
My main coding environment right now is a VMware instance with 128GB of RAM, 32 virtual processors running on top of a single i9-14900K, and 2TB of disk space. I will likely move to a dedicated box in the near future.
My local LLM works off of one Blackwell RTX 6000 that I rent from a Canadian company, ExducerAI. I may bring that in locally in the future, but not today.
What models do I use?
For hosted models, I use both Claude Code and Codex. I find that, depending on the month, one is better than the other. I tend to swap between them both fairly regularly, or code in one and refactor / QA in the other. I have not yet gone ham with openrouter, umans or similar things but I suspect that is in my future.
For local models, I use Qwen-coder for building and Command A+ for reasoning. These answers change quite often, given the speed at which SOTA models come out.
I try and support Canadian models where possible, given the sovereignty discussions that I have been having around Canada.
What's my harness?
Claude Code and Codex work really well. For all local considerations, I use Hermes. The word harness seems to have a lot of different meanings to different people. I'm not sure if I'm using the term in the right way.
My AI engine constantly creates things I don't want it to
I rely on Superpowers to focus my AI engines. I can't AI without it. It is core to my agentic setup.
What's in your .md files?
My AGENTS.md and CODEX.md are heavily influenced by Jesse and Harper. I heard YADM is neat, but I haven't tried it yet.
What if I want a UI? CLI sucks!
I use VS Code when I want one, but I barely use it. CLI is great and where I live.
How do you move .md files back and forth?
I'm still looking for a better way to do this. I currently just use WinSCP on the box, but it sucks. VS Code can open .md files remotely, but I find it slow and very heavy. Obsidian might help me here, but I haven't figured out how yet.
What if I want to access my CLIs from anywhere?
I use Termius on my phone, Windows box and Mac laptop to ssh into my server. Tailscale is what lets me effortlessly access the server. I use Herdr on the server to manage my various sessions. This essentially gives me access to all my projects anywhere in the world on any device.
I want to track my usage
I use AgentsView to understand my usage patterns.
I want AI-generated code to be reviewed in an automated way
I want multiple agents to work on the same codebase
Worktrunk is the way.
How do you handle security so that passwords aren't leaking everywhere?
fnox works for me.
AI doesn't write well
Jesse's Elements of Style plugin is pretty rad.