I cancelled my Claude subscription today because I can't use my purchased tokens in my harness of choice. I also couldn't use them with an LLM proxy server, either. Sure, I can do some hacky things to make it work, but it's not supported and I risk getting banned permanently if they decide that I've abused the terms of service.
I'm not loyal to any provider or service. I've tried Qwen, MiniMax, Xiaomi, OpenAI, Anthropic, OpenRouter, etc. I set up my proxy server so swapping between providers is easy on the harness side, I just toggle /model and choose. I've also configured an autorouting "model" in the proxy which helps move the outgoing request to the right model+thinking+provider combination I want.
The global frontier models from the last four to six months have gone from that's pretty cool to wow, holy shit, this is the best way to create software. But no model has a monopoly (yet). The only publicly available harness that has kept pace is pi, best I can tell. They just released pi-durable too.
If you're gonna lock me into your harness, make it worth my fucking time (and therefore money). Double my subscription credit while I'm using your harness, provide me better inference speed than outside the harness, don't nanny my messages by downgrading the model to a lower tier, allow me to try beta models with no data retention, etc.
Before long, I will have a completely custom harness for daily development. It's close now, but I still enter pi into the terminal before making incredible things.