I agree that the comparison between langchain/llamaindex is probably the better one.
With that being said, txtai has a much more in-depth approach with how it builds it's data stores vs just assuming the underlying systems will handle everything. It supports running SQL statements and integrates the components in a way other RAG systems don't. It was also a vector store before it had a RAG workflow. There are years of code behind that part.