SERVER / VECTOR
Find related records
Store embeddings, attach JSON metadata, and find related records with the same server-only contract in local and remote environments.
Cache documentation
1 / CONNECTION
Declare Vector, operate with emb
The initial provider is dowe. Connection values stay server-only and may come from the server environment.
vector appVector provider:"dowe" host:env.VECTOR_HOST port:env.VECTOR_PORT account:env.VECTOR_USER secret:env.VECTOR_PASSWORD name:env.VECTOR_DATABASEField | Meaning |
|---|---|
No dataThere are no records to display | |
2 / OPERATIONS
Upsert, search, read, delete, and list
The first upsert fixes the dimension. Search uses cosine similarity and returns the highest scores first.
emb saved conn:appVector.upsert id:"article-1" vector:[0.12 0.48 -0.2] metadata:{ kind:"article" }
emb matches conn:appVector.search vector:req.body.vector limit:10 minScore:0.7 where:{ kind:"article" }
emb article conn:appVector.read id:req.params.id required:true
emb removed conn:appVector.delete id:req.params.id
emb articles conn:appVector.list limit:100 where:{ kind:"article" }Operation | Contract |
|---|---|
No dataThere are no records to display | |
3 / LOCAL AND REMOTE
One contract, explicit production mode
The same declarations work during development and production. The host value decides whether the connection stays embedded or crosses an authenticated WebSocket.
Mode | Behavior |
|---|---|
No dataThere are no records to display | |
4 / DOWE VECTOR
Run the remote provider
The maintained dowe-vector project authenticates an account and bearer secret before the WebSocket upgrade.
main
server port:4149
vector servicedowe vector create-account semantic-search search-api