sonamu-vector
Installation
SKILL.md
Vector Search Guide
sonamu/vector exports embedding, chunking, configuration, and result types. It does not export a
search service: store vectors through the consuming project's Model and compose pgvector queries
with Puri from "sonamu".
Structure
| Task | Read |
|---|---|
Install a provider, configure its key, call Embedding.embedOne() / embed(), or diagnose provider, batching, token-count, or dimension behavior |
references/embeddings.md |
Split text, tune Chunking, interpret offsets and overlap, or decide whether token-aware splitting is needed |
references/chunking.md |
Declare and persist a vector column, add an index, call Puri.vectorSimilarity(), or combine vector and text search |
references/persistence-and-search.md |
Start at the text boundary: use "document" while generating stored Voyage vectors and "query"
while generating their search vectors. Then keep the provider/model and vector dimensions aligned
between both sides and the entity column.