The best foundation for agents is Postgres®. The best Postgres is EDB Postgres AI.
McKnight Consulting Group benchmarks put EDB Postgres AI ahead of dedicated vector databases, lakehouse platforms, and every other tested Postgres provider on speed, accuracy, and cost for production agents.
Retrieval that lives in the database, not beside it.
A separate vector database adds another system, serves stale data, and can't do hybrid search. EDB Postgres AI runs vector and hybrid search on data that's always current, right in the database you already run—better than the competition:
Validated performance
McKnight Consulting Group, July 2026
At 50 million vectors, all platforms normalized to equivalent enterprise hardware (512 GB memory, 64–72 vCPU, us-west-2):
| Platform | Median query latency (p50) | Recall@10 | Effective $/hr (speed-adjusted) |
|---|---|---|---|
| EDB Postgres AI | 50 ms | 0.884 | $16.97 |
| Crunchy Bridge | 98 ms | 0.868 | $20.89 |
| AWS Aurora | 95 ms | 0.871 | $21.81 |
| MongoDB Atlas | 1,083 ms | 0.693 | $577.46 |
| Databricks | 4,023 ms | 0.738 | $1,287.36 |
New writes are queryable in 12 ms, the fastest write-to-read of any platform tested, so agents always act on live data.
Open source PostgreSQL was also tested (78 ms, 0.873 recall at 50M); EDB Postgres AI still leads it on both latency and accuracy. Effective $/hour normalizes each platform's as-tested hourly rate by how many times slower it runs than EDB Postgres AI at 50M vectors. Full methodology in the report.
"Across every workload—vector search, hybrid queries, and full end-to-end agent retrieval—EDB Postgres AI led the field, and did so while spending fewer tokens and dollars to get there. It was the strongest implementation we evaluated." — McKnight Consulting Group, July 2026