RagLeapPluggable vector backends beyond ragleap-rag's built-in 6 (PgVector, FAISS, Pinecone, Weaviate, Qdrant, Milvus).
This package is under active development - more backends will be added over time. See the roadmap for status.
Available backends
| Backend | Extra | Notes |
|---|---|---|
| Chroma | chroma |
Embedded/local via chromadb's PersistentClient - no server required. No native sparse/keyword search (supports_sparse() is False); hybrid search falls back to dense-only. |
| LanceDB | lancedb |
Embedded/local via a directory path - no server required. Real upsert semantics via merge_insert(). No native sparse/keyword search enabled yet (supports_sparse() is False); hybrid search falls back to dense-only. |
| Redis | redis |
Requires a real running server - Redis Stack, or plain Redis with the RediSearch module loaded (no embedded/local mode). init_schema() checks for the module and raises a clear error if it's missing. Metadata filtering only supports document_id (RediSearch requires predeclared schema fields). No native sparse/keyword search enabled yet (supports_sparse() is False); hybrid search falls back to dense-only. |
| Upstash Vector | upstash |
Managed serverless REST API - no embedded/local mode at all, and the index (fixed dimension, dense-vs-hybrid type) must already exist, created via the Upstash console; init_schema() verifies compatibility via info() rather than creating anything. Metadata filtering supports arbitrary multi-key filters natively (real SQL-like filter= string over a genuine JSON dict) - more flexible than Redis here. No native sparse/keyword search enabled yet (supports_sparse() is False); hybrid search falls back to dense-only. |
| OpenSearch | opensearch |
Requires a real running OpenSearch instance (no embedded/local mode); uses OpenSearch's native k-NN vector search with the lucene engine (the nmslib engine is rejected on OpenSearch 3.0+). The document registry is a second OpenSearch index. Metadata filtering supports arbitrary term/range filters over any mapped field. No native sparse/keyword search enabled yet (supports_sparse() is False); hybrid search falls back to dense-only. |
Design
Every backend here implements ragleap-rag's VectorBackend interface, so it
can be passed directly to RagLeap(vector_backend=...). Each backend's real
client SDK is an optional extra - installing ragleap-vectorstores alone
pulls in no heavy dependencies beyond ragleap-rag itself.
Install
pip install ragleap-vectorstores[chroma] # or, with uv uv add ragleap-vectorstores[chroma] pip install ragleap-vectorstores[lancedb] # or, with uv uv add ragleap-vectorstores[lancedb] pip install ragleap-vectorstores[opensearch] # or, with uv uv add ragleap-vectorstores[opensearch]
Usage
from ragleap_vectorstores import ChromaBackend backend = ChromaBackend(persist_directory="./chroma_data")
from ragleap_vectorstores import LanceDBBackend backend = LanceDBBackend(uri="./lancedb_data")
from ragleap_vectorstores import RedisBackend # Requires a real Redis Stack instance (or plain Redis + the RediSearch # module) - plain Redis alone has no vector search. backend = RedisBackend(redis_url="redis://localhost:6379/0")
from ragleap_vectorstores import UpstashBackend # Requires a real Upstash Vector index, created via the Upstash console # (console.upstash.com) as a pure DENSE index - no embedded/local mode. backend = UpstashBackend(url="https://...upstash.io", token="...")
from ragleap_vectorstores import OpenSearchBackend
# Requires a real running OpenSearch instance (e.g. http://localhost:9200) -
# no embedded/local mode. Pass http_auth=("user", "password") and
# use_ssl=True for a secured cluster.
backend = OpenSearchBackend(opensearch_url="http://localhost:9200")