Results
Run a search to see matching papers.
How it works
One embedding request. One local vector search. Metadata filters are applied before wick returns the nearest records.
- Engine
- C++20 Persistent local storage
- Index
- HNSW Cosine distance
- Embedding
- all-MiniLM-L6-v2 384 dimensions
- Corpus
- 18 original papers
Benchmarks
wick records retrieval quality and system cost together, using repeatable local workloads.
- Quality
- Recall@k checked against independent exact results
- Speed
- P50, P95, P99, QPS, and insertion throughput
- System
- Memory, disk use, startup time, and recovery time