langchain-opensolr

LangChain integration for Opensolr — managed Apache Solr with server-side embeddings and native hybrid (BM25 + kNN) search.

Product page: opensolr.com/langchain · Platform: opensolr.com (managed Solr hosting since 2011 — free 15-day trial, no card)

No local embedding model. No third-party embedding API key. One set of credentials, and the vectors are computed on Opensolr's GPU infrastructure (multilingual E5-large-instruct, 1024 dimensions, cosine).

pip install langchain-opensolr

The whole tutorial

from langchain_opensolr import OpensolrVectorStore

vs = OpensolrVectorStore(
    index="mysite__dense",          # vector-enabled Opensolr index
    email="you@example.com",
    api_key="YOUR_OPENSOLR_API_KEY",
    create_if_missing=True,          # provisions the index on first use
)

vs.add_texts(
    ["Hybrid search fuses BM25 with vector similarity",
     "Cats sleep sixteen hours a day"],
    metadatas=[{"category": "search"}, {"category": "animals"}],
)

docs = vs.similarity_search("how do lexical and semantic search combine?", k=1)
print(docs[0].page_content)

That's it — no embedding model was configured, because embedding happens on the server at both index and query time.

Hybrid search

Pure vector search fails on exact identifiers; pure BM25 fails on meaning. Opensolr's {!hybrid} query parser fuses both scores per document:

docs = vs.similarity_search(
    "affordable restaurants",
    k=5,
    hybrid=True,
    mode="union",     # union | keywords_required | meaning_required | intersection
    alpha=0.5,        # 0 = all semantic … 1 = all lexical
)

Metadata filters

vs.similarity_search("search engines", k=5, filter={"category": "search"})
vs.similarity_search("anything", k=5, filter='meta_rank:[2 TO *]')   # raw Solr fq

Metadata round-trips losslessly (stored as JSON alongside filterable meta_* fields).

As a retriever, in any chain

retriever = vs.as_retriever(search_kwargs={"k": 5, "hybrid": True})

Standalone embeddings

Use Opensolr's embedding endpoint with any other LangChain component:

from langchain_opensolr import OpensolrEmbeddings

emb = OpensolrEmbeddings(email="you@example.com", api_key="...", index="mysite__dense")
emb.embed_query("budget-friendly dining")   # -> 1024 floats

Notes

• Vector-enabled indexes run on Opensolr's Solr 9.x environments — currently us (Chicago), de (Germany), fi (Finland). Pass location= to choose. The list is fetched live from the platform, so new regions work without a package upgrade — and additional dedicated regions can be deployed on request (paid add-on): support@opensolr.com.
• A free Opensolr account (15-day trial, no card) includes an AI quota that comfortably covers this README end to end: opensolr.com.
• Full platform docs: AI & Vector Search.

Development

pip install -e . pytest
pytest tests/unit_tests
OPENSOLR_EMAIL=... OPENSOLR_API_KEY=... OPENSOLR_INDEX=... pytest tests/integration_tests

MIT license.