Claude has been down on and off this afternoon. It took me about 20 minutes to port over my skills, tasks, and memory to Codex. The code was already in GitHub. My memory files were either stored locally or exported from Anthropic (something I had done previously). The switch was surprisingly straightforward. What struck me wasn't that I switched—it was how quickly I stopped caring. Within minutes, the AI tool I couldn't imagine working without just hours ago had been replaced because I simply needed to keep moving. That's how brutal is this market!?
Mengmeng Chen this is the real benefit of keeping the workspace and files backed up and local. You don’t have to rebuild your context from scratch when you switch tools. You’re just changing the interface around the same projects, instructions, and memory. I switch between Claude and Codex regularly. Claude still has a lot of features I value, but Codex has recently become my daily driver. Being able to keep working from the same local workspace makes switching feel practical instead of disruptive. Hope you enjoy Codex!
smart. You've built a model agnostic stack. I'm doing same. My backup stack involves Hermes agent. I'm less of a coder and more involved in knowledge work otherwise I'd be using Codex. Glad to see someone else planning ahead and building in redundancy.
You can actually use Supermux ADE to work in an agent agnostic way. It supports Claude Code, Codex, OpenCode, Grok Build and Cursor. The agents use shared memory and skills.
Twenty minutes is a lot shorter than I'd have guessed for that kind of switch. Good argument for keeping the memory files somewhere you own.
If you were running Claude in orca you could move the sessions in 1 click. Such an upgrade 🙏🏻
When the product is intelligence, there is nothing stopping you from one "intelligence-commodity" provider to another, especially to open-weight and open-source models. One cannot help but wonder about the end-goal of these frontier labs.