Storage Backends

Chengeta AI provides nine interchangeable backends — four key-value stores, one tiered backend, one async backend, and three vector similarity stores. Every backend conforms to a Protocol, so you can swap implementations without changing application code.

---

Key-Value Backends (CacheBackend)

Backend	Class	Extras	Best For
In-Memory (LRU)	InMemoryBackend	— (core)	Dev, testing, single-process
Async In-Memory	AsyncInMemoryBackend	— (core)	FastAPI, async LangGraph
Tiered (L1+L2)	TieredBackend	— (core)	Memory speed + Redis persistence
Disk	DiskBackend	— (core)	Single-node persistence
Redis	RedisBackend	[redis]	Shared across processes / services

Vector Backends (VectorBackend)

Backend	Class	Extras	Best For
FAISS	FAISSBackend	[vector-faiss]	Fast in-process similarity
ChromaDB	ChromaBackend	[vector-chroma]	Persistent vector store + metadata
Qdrant	QdrantBackend	[vector-qdrant]	Production scale (22ms p95)
Weaviate	WeaviateBackend	[vector-weaviate]	Hybrid search (vector + BM25)

---

Decision Flowchart

---

Protocols

Both protocols are @runtime_checkable — verify with isinstance().

CacheBackend

from chengeta_ai.backends.base import CacheBackend
# get / set / delete / exists / clear / close

AsyncCacheBackend

from chengeta_ai.backends.async_base import AsyncCacheBackend
# async get / set / delete / exists / clear / close

VectorBackend

from chengeta_ai.backends.base import VectorBackend
# add / search / delete / clear / close

---

Custom Backend

Implement the CacheBackend protocol — no inheritance required:

class MyBackend:
    def get(self, key: str): ...
    def set(self, key: str, value, ttl=None): ...
    def delete(self, key: str): ...
    def exists(self, key: str) -> bool: ...
    def clear(self): ...
    def close(self): ...