Right now the Ask flow waits for the full answer + verifier pass before rendering. UX wins if we stream tokens AS the verifier confirms each sentence — users see grounded sentences appear progressively, with a 'verifying...' indicator on the last sentence.
Plan:
- Switch backend to SSE on
/v1/ask/stream
- Server emits per-sentence events with chunk citations attached
- Frontend renders sentence-by-sentence with citation chips appearing inline
Right now the Ask flow waits for the full answer + verifier pass before rendering. UX wins if we stream tokens AS the verifier confirms each sentence — users see grounded sentences appear progressively, with a 'verifying...' indicator on the last sentence.
Plan:
/v1/ask/stream