Ellipsis uses BUNCH of LLM agents internally. If you built your own code generation LLM agent you'd need to also build a way to execute the code that the agent writes, which is a bit of an engineering headache.

We handle this, the result is that if you set up a Dockerfile, we promise to return working, tested code: https://docs.ellipsis.dev/code#from-a-pr