KiviDB case studies: RAG retrieval, semantic caching, agent memory and CDC, measured
KiviDB Cloud measured on public benchmarks: retrieval quality and latency on BEIR SciFact, semantic caching on Banking77, agent memory on LoCoMo, and PostgreSQL change data capture.
Each study runs a common AI or data workload on KiviDB Cloud against a dataset the field already uses to compare systems. We publish the method, every number, and what did not work.
Read a study
- RAG retrieval on SciFact: Keyword, vector and hybrid search scored on 300 scientific claims.
- Semantic cache on Banking77: 13,083 support questions: model calls saved against wrong answers.
- Agent memory on LoCoMo: 1,540 questions from retrieved memories against the whole history.
- PostgreSQL to KiviDB, live: Change data capture timed from commit to readable, and a fix it led to.
How we run them
- Public data only Datasets anyone can download, so every number can be checked. No customer data and no customer names.
- A free database Each study runs on a free-tier KiviDB Cloud database. Where hardware matters, we say so and say where the timing was taken from.
- Everything reported Results that did not favour us stay in. Each study lists the model, the parameters and the measures used.
Run it on your data
Create a free database and try the same workload with your own documents, prompts or tables.