The text2vec-transformers Docker service was removed in Jan 2026, but retrieval_tools.py still used near_text() which requires it. Now uses GPU embedder (BGE-M3) with near_vector() like flask_app.py. Changes: - Add GPU embedder singleton (get_gpu_embedder) - search_chunks_handler: near_text → near_vector + BGE-M3 - search_summaries_handler: near_text → near_vector + BGE-M3 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>