Trieve’s infrastructure combines Postgres and a dedicated search engine, presenting challenges in terms of scalability and performance. The use of pgvector with Postgres may not be a complete solution due to limitations in handling dense vector searches and query requirements including negated words. Notably, Postgres lacks the ability to highlight matched keywords, which affects user experience. Additionally, pgvector can be slow on filter and order by queries, requiring continuous tuning. While PGVector offers semantic search capabilities, it may not fully replace traditional full-text search methods. Trieve, with its expertise, addresses these issues effectively. It’s important to consider the trade-offs between simplicity and functionality when choosing a search solution.
https://trieve.ai/pgvector-missing-features