RAG on managed Apache Solr — one pip install

Try it without an account

A public demo account, so the quickstart above runs before you decide anything.

export OPENSOLR_EMAIL=mcp@opensolr.com
export OPENSOLR_API_KEY=420b8b23e7b12dc8ab838932145a5065

The index mcp_demo_d1__dense is already loaded with 300 news articles, so search, filtering and grounded answers work the moment you connect. You also get the write path: create your own index on the account, ingest into it and query it. Deletion is not available on this shared key — no index here can be deleted or reconfigured by hand. Whatever you create is removed automatically after 3 days.

• Anything you create there is deleted after 3 days. Automatically, without warning or export — indexes you created included.
• The account is shared with everyone reading this page. Your index is visible to them and they can add documents to it, as you can to theirs. Nobody can delete or reconfigure an index here — that is switched off for this key — but never put anything real, private or client-owned in it.
• The limits are per index, and deliberately small. 200 MB of bandwidth and 50 MB of disk per index. Bandwidth is the one you will hit first: it covers a demo, a tutorial and a proof of concept, and it will not carry an application.

When you want an index that is private, yours and still there next week, get your own key — the free plan is free forever, no card — and change the two variables above. Nothing else in your code changes.

What happens under the hood

Your app talks to one package. The package talks to a managed pipeline that already exists.

Two lines you'll actually use

Hybrid retrieval with metadata filters, and the retriever that plugs into every RAG example ever written.

docs = vs.similarity_search(
    "affordable restaurants",
    k=5,
    hybrid=True,      # BM25 + kNN, fused
    mode="union",     # or keywords_required,
                      # meaning_required,
                      # intersection
    alpha=0.5,        # semantic ↔ lexical
    filter={"city": "Cluj"},
)

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

# now use it anywhere LangChain
# expects a retriever:
chain = (
    {"context": retriever, "question": ...}
    | prompt | llm
)

Local files, PDFs page by page

Each file is read on Opensolr’s servers — its text, the text in its pictures (OCR, more than 100 languages), a short description of the pictures without text — and a PDF becomes one document per page, so a search lands on the page that answers it instead of a 200-page file that mentions it somewhere. Word, RTF, OpenDocument, text, HTML and scanned pictures are one document each. In search results each PDF then comes back once, as its best page, with a link that opens it right there.

from langchain_opensolr import OpensolrLoader

# Files read on Opensolr: one Document per PDF page, scans through OCR
loader = OpensolrLoader("./contracts", index="mysite__dense",
                        email="you@example.com", api_key="...", pattern="*.pdf")
docs = loader.load()      # metadata: source, page, is_ocr, language

# Or straight into the store — no splitter, PDFs page by page:
vs.add_files("./contracts", pattern="*.pdf", metadata={"team": "legal"}, wait=True)

# A PDF comes back as its best page, with a link to it; a period narrows the search
docs = vs.similarity_search("termination notice period", k=4, hybrid=True,
                            date_from="2025-01-01")
docs[0].metadata["page_url"]   # https://.../msa.pdf#page=14  (+ more_pages, document_id)

The bytes travel once: a file the account already sent is not uploaded again, and sending the same files again replaces their documents. Pictures read by OCR count 0.5 of an AI request each; a file read before is free. Big files are read in the background: whatever is still being read comes back as pending — run again to finish them.

Your Search Tuning follows you

Relevance is configured once, in your Control Panel — not re-implemented in every codebase.

• Saved per index: everything you set in Index Settings → Search Tuning (semantic↔lexical balance, field weights, minimum match, search mode, vector candidate pool, content quality boost) is stored with your index.
• Applies automatically here: every search and every RAG answer from this integration runs through the same tuned pipeline as your hosted search page — change a slider in the Control Panel and the very next query uses it. No redeploy, no code change.
• Overridable per call: tuning={"search_mode": "keywords_required", "fw_title": 0.2, "mm": "strict"} beats the saved settings for that one request. Defaults match the platform exactly when you send nothing.
• Fresh Results Bias: pass fresh_bias to rank newer documents higher — scores are multiplied by a recency curve on creation_date. It re-orders and never filters: the hit count is unchanged and undated documents keep their place. This is the same control visitors get as the Fresh toggle on the hosted search page. Not to be confused with freshness_boost, which is a hard date window in days and does remove results.