How to create a Qdrant document store in Haystack

A retrieval application needs its indexed documents to survive beyond one Python process. QdrantDocumentStore connects Haystack to Qdrant, where embeddings and metadata remain available for later semantic searches.

The qdrant-haystack integration supports a disk-backed local mode through the path parameter. Local mode keeps this smoke test self-contained while exercising the same document-store and QdrantEmbeddingRetriever handoff used with a Qdrant server.

The sample uses three-dimensional vectors so it needs neither an embedding model nor an API key. A production index must use the dimension produced by its document and query embedders, and recreate_index=True must be removed before opening a collection whose existing documents must be preserved.

1. Install the qdrant-haystack integration beside haystack-ai in the Python environment that runs the application.

   $ python -m pip install qdrant-haystack

   Related: How to install Haystack with pip

2. Create qdrant_store_demo.py with the imports and disk-backed document store.

   qdrant_store_demo.py

   from pathlib import Path
    
   from haystack import Document
   from haystack_integrations.components.retrievers.qdrant import QdrantEmbeddingRetriever
   from haystack_integrations.document_stores.qdrant import QdrantDocumentStore
    
    
   document_store = QdrantDocumentStore(
       path="support-qdrant-demo",
       index="support_articles",
       embedding_dim=3,
       similarity="cosine",
       recreate_index=True,
       return_embedding=True,
       progress_bar=False,
   )

   recreate_index=True replaces an existing collection with the same index name. Using this value with a retained collection deletes its indexed documents.

3. Append the three embedded support documents below the QdrantDocumentStore block.

   documents = [
       Document(
           content="Reset passwords through the identity portal.",
           meta={"topic": "accounts"},
           embedding=[0.99, 0.03, 0.01],
       ),
       Document(
           content="Restart the search service after changing the index.",
           meta={"topic": "operations"},
           embedding=[0.02, 0.98, 0.02],
       ),
       Document(
           content="Archive old invoices from the billing dashboard.",
           meta={"topic": "billing"},
           embedding=[0.01, 0.02, 0.99],
       ),
   ]

   Each embedding has the three values required by embedding_dim=3, and the vectors point in different directions so the retrieval result is unambiguous.

4. Append the document write and embedding retrieval below the document list.

   written = document_store.write_documents(documents)
    
   retriever = QdrantEmbeddingRetriever(
       document_store=document_store,
       top_k=1,
       return_embedding=True,
   )
   match = retriever.run(query_embedding=[1.0, 0.0, 0.0])["documents"][0]

5. Append the stored-state and retrieval checks below the match assignment.

   print(f"documents written: {written}")
   print(f"documents stored: {document_store.count_documents()}")
   print(f"top match: {match.content}")
   print(f"match topic: {match.meta['topic']}")
   print(f"embedding returned: {match.embedding is not None}")
   print(f"qdrant path exists: {Path('support-qdrant-demo').is_dir()}")

6. Confirm the assembled qdrant_store_demo.py matches the complete program before execution.

   qdrant_store_demo.py

   from pathlib import Path
    
   from haystack import Document
   from haystack_integrations.components.retrievers.qdrant import QdrantEmbeddingRetriever
   from haystack_integrations.document_stores.qdrant import QdrantDocumentStore
    
    
   document_store = QdrantDocumentStore(
       path="support-qdrant-demo",
       index="support_articles",
       embedding_dim=3,
       similarity="cosine",
       recreate_index=True,
       return_embedding=True,
       progress_bar=False,
   )
    
   documents = [
       Document(
           content="Reset passwords through the identity portal.",
           meta={"topic": "accounts"},
           embedding=[0.99, 0.03, 0.01],
       ),
       Document(
           content="Restart the search service after changing the index.",
           meta={"topic": "operations"},
           embedding=[0.02, 0.98, 0.02],
       ),
       Document(
           content="Archive old invoices from the billing dashboard.",
           meta={"topic": "billing"},
           embedding=[0.01, 0.02, 0.99],
       ),
   ]
    
   written = document_store.write_documents(documents)
    
   retriever = QdrantEmbeddingRetriever(
       document_store=document_store,
       top_k=1,
       return_embedding=True,
   )
   match = retriever.run(query_embedding=[1.0, 0.0, 0.0])["documents"][0]
    
   print(f"documents written: {written}")
   print(f"documents stored: {document_store.count_documents()}")
   print(f"top match: {match.content}")
   print(f"match topic: {match.meta['topic']}")
   print(f"embedding returned: {match.embedding is not None}")
   print(f"qdrant path exists: {Path('support-qdrant-demo').is_dir()}")

7. Run the completed smoke test from the directory that will retain the local Qdrant data.

   $ python qdrant_store_demo.py
   documents written: 3
   documents stored: 3
   top match: Reset passwords through the identity portal.
   match topic: accounts
   embedding returned: True
   qdrant path exists: True

   The write and stored counts confirm the collection contains all three documents. The matching account article, returned embedding, and persistent directory confirm that QdrantEmbeddingRetriever can read the created store.