It kept truncating files only about 600 lines long. It also seems to rewrite the entire file each time instead of just sending diffs like aider making it super slow.
The 600 line limit might be due to the output token limit on the LLM (not sure what they are using for the code rewriting)