PineconeBackend

Pinecone serverless vector backend for semantic caching at cloud scale. Creates the Pinecone index automatically if it doesn't exist.

Installation

pip install 'chengeta-ai[vector-pinecone]'

---

Usage

Basic setup with SemanticCache

from chengeta_ai.backends.pinecone_backend import PineconeBackend
from chengeta_ai import SemanticCache, InMemoryBackend

vector_backend = PineconeBackend(
    api_key="pc-your-api-key",
    index_name="chengeta-semantic",
    dim=1536,
)

cache = SemanticCache(
    exact_backend=InMemoryBackend(),
    vector_backend=vector_backend,
    embed_fn=lambda text: embed_model.encode(text),
    threshold=0.92,
)

cache.set("What is the capital of France?", "Paris")
result = cache.get("France's capital city?")  # "Paris"

Custom cloud and region

vector_backend = PineconeBackend(
    api_key="pc-your-api-key",
    index_name="chengeta-eu",
    dim=1536,
    cloud="gcp",
    region="europe-west1",
)

With AdaptiveSemanticCache

from chengeta_ai.layers.adaptive_semantic_cache import AdaptiveSemanticCache
from chengeta_ai.backends.redis_backend import RedisBackend

cache = AdaptiveSemanticCache(
    exact_backend=RedisBackend(url="redis://localhost:6379/0"),
    vector_backend=PineconeBackend(
        api_key="pc-...",
        index_name="chengeta",
        dim=1536,
    ),
    embed_fn=embed_fn,
    target_hit_rate=0.35,
)

Auto index creation

If index_name doesn't exist in your Pinecone account, the backend creates it automatically using the specified cloud and region with cosine metric.

Serverless vs dedicated

Pinecone serverless scales to zero — no idle cost. Use serverless for dev/staging. For production SLAs requiring consistent p99 latency, consider Qdrant or a dedicated Pinecone pod.

---

API Reference

Constructor

Parameter	Type	Default	Description
api_key	str	required	Pinecone API key
index_name	str	required	Index name (created if absent)
dim	int	required	Embedding dimension
cloud	str	"aws"	Cloud provider for index creation
region	str	"us-east-1"	Region for index creation
top_k	int	1	Default number of candidates to fetch

Methods

Method	Description
add(key, vector, metadata)	Upsert a vector with key stored as metadata
search(vector, top_k)	Return (key, score) pairs for nearest neighbours
delete(key)	Remove a vector by key
clear()	Delete all vectors (delete_all=True)
close()	No-op (Pinecone client is stateless)