Comment by contingencies

So... Multi-target = {chat, web, markdown}; Audience = {human, bot}; USPs = {speed, unified model}

Since I don't believe the maintenance challenges of a non-unified model are particularly high-friction in the LLM era (even if they may be error-prone, since that is usually solved with CI/CD or TDD), the nominal utility of this angle of architectural enhancement is considered dubious.

With regard to speed, there is no hardware shortage at present for any of these simple rendering tasks.

Overall, a nice efficient codebase but suffering a bit from NIH syndrome solving no particularly acute problem. Unless a community forms rapidly, this will probably die.

I'd like to see the benchmark compared to, eg. static files served with binserve.