A: We actually did consider adding OpenRouter. There are two issues with it. Allowing just any model means that some tasks will fail and will keep failing. This would create a really bad user experience. Second, we cannot trust unknown or less known providers to not leak our prompts and workflow logic.

The only way we could add OpenRouter would be if we were able to vet every model in advance and provide an allowed models list. However, this is a very hefty task and we can really tackle it at our current scale. To put things in perspective, big companies have a hard time setting up long-running workflows with a single LLM provider - we manage to handle three providers across days, but there is a limit to how much we can manage too.