Modem moves vector search from pgvector to turbopuffer, cutting p90 latency 76%

Modem moved its embeddings out of a shared multi-tenant Postgres pgvector table and into turbopuffer, a paid proprietary search database. Modem says the switch cut p90 search time by 76% and took most of the load off its database. The shared pgvector table had let the smallest customers wait up to ten seconds, and adding an HNSW index made performance worse because vector search ignored the organization_id and time filters. By August 2026 the median search took about 0.8 seconds and the slowest 1% took as long as 28 seconds. Modem says that even with turbopuffer's fees, the dual-system setup lowered its entire storage bill, covering all Modem data and not just embeddings, by roughly 10%.

Generated from reports · updated 1 h ago

Not enough continuous observations yet to draw a trend.