WeaviateBackend

Weaviate-based vector similarity backend. The only vector database with native hybrid search (semantic + BM25 keyword) without a separate layer. Supports embedded (local) and Weaviate Cloud instances.

Installation

pip install 'omnicache-ai[vector-weaviate]'

---

Usage

Embedded (local, no server)

from omnicache_ai.backends.vector_backend import WeaviateBackend
from omnicache_ai import SemanticCache, InMemoryBackend

vector_backend = WeaviateBackend()  # embedded local instance

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

Weaviate Cloud

vector_backend = WeaviateBackend(
    url="https://your-cluster.weaviate.network",
    api_key="your-weaviate-api-key",
    class_name="OmnicacheEntry",
)

---

Why Weaviate?

Weaviate is the only vector database with native hybrid search — it combines BM25 keyword ranking with vector similarity in a single query without a separate BM25 layer. This makes it ideal for RAG caches where queries mix keyword specificity (e.g. product codes) with semantic meaning.

Feature	Weaviate	Qdrant	FAISS
Native hybrid search	✅	❌	❌
Persistent	✅	✅	❌
Cloud-managed	✅	✅	❌
Embedded (no server)	✅	❌	✅

---

API Reference

Parameter	Type	Default	Description
url	str | None	None	Weaviate instance URL. None = embedded
api_key	str | None	None	Weaviate Cloud API key
class_name	str	"OmnicacheEntry"	Weaviate class name for cache entries