Pinecone¶
Serverless hosted vector store. Vectors in Pinecone, RL scores in a local SQLite sidecar.
Install¶
Create an index¶
from pinecone import Pinecone, ServerlessSpec
pc = Pinecone(api_key="pk_...")
pc.create_index(
"extremis",
dimension=384, # match EXTREMIS_EMBEDDING_DIM
metric="cosine",
spec=ServerlessSpec(cloud="aws", region="us-east-1"),
)
Configure¶
RL scores¶
Pinecone doesn't support partial metadata updates efficiently. extremis stores RL scores in a local SQLite file alongside the Pinecone index:
Change the path: EXTREMIS_PINECONE_SCORE_DB=/custom/path.db
Namespace mapping¶
Pinecone namespaces map 1:1 to extremis namespaces.