RAG vs raw

One question, two answers, side by side. The difference is the retrieval.

The same question answered with and without retrieval.

What is real here

The right-hand column is a real call to this site's assistant: real retrieval over the real corpus, plus the site's own records for questions about his work as a whole, streamed token by token, with the token counts and the cost the server itself reports. The left-hand column is honest about a gap — a raw, ungrounded answer would need an endpoint this API deliberately does not publish.

All demos