I was thinking about why CC isn't opensource and forkable, and then I realized it's probably mostly about optimizing the costs of inference on Anthropic's side and caching sysprompts etc properly.
They built their infra around that. It's hard to blame them for trying to save money on compute for subscriptions.