We shipped v2 of serverlessVector – a pure-Go, in-memory vector DB aimed at serverless and small/medium datasets (no ANN, linear search).
v2 highlights
- float32 only – matches OpenAI/Cohere/sentence-transformers;
Vector.Data is []float32, no more float64/type switching.
- SearchMMR – Maximal Marginal Relevance (relevance + diversity), plus
SelectMMRFromCandidates for re-ranking and SearchMMRWithScores for blending with external scores (e.g. keyword).
- SearchWithFilter – filter by metadata/tags at query time.
- Faster – float32 storage, heap topK, shorter locks in BatchAdd/GetStats.
Install: go get github.com/takara-ai/serverlessVector/v2@latest
Docs and benchmarks: https://github.com/takara-ai/serverlessVector
We shipped v2 of serverlessVector – a pure-Go, in-memory vector DB aimed at serverless and small/medium datasets (no ANN, linear search).
v2 highlights
Vector.Datais[]float32, no more float64/type switching.SelectMMRFromCandidatesfor re-ranking andSearchMMRWithScoresfor blending with external scores (e.g. keyword).Install:
go get github.com/takara-ai/serverlessVector/v2@latestDocs and benchmarks: https://github.com/takara-ai/serverlessVector