Qdrant Vector Database

The project uses Qdrant for retrieval in the answer pipeline and for indexing Zammad knowledge base content.

Integration

The workflow service configures Qdrant under answer.qdrant. The index job configures it under qdrant.

url: The URL of the Qdrant instance.
api_key: Secret key for authentication.
collection_name: The name of the collection where knowledge vectors are stored.
vector_dimension: The dimensionality of the embeddings.
vector_name: Optional name of the vector configuration in the collection.
retrieval_num_documents: Number of documents to retrieve per query.
retrieval_mode: dense, sparse, or hybrid.
sparse_vector_name: Name of the sparse vector configuration.
multi_query.enabled: Enable multi-query expansion.
multi_query.include_original: Keep the original query in the retrieval set.

The workflow stores answer documents with title and url fields. The index job stores knowledge base and law metadata in Qdrant payloads.

Knowledge base entries are compared by content hash before being written.
Law entries use deterministic IDs and store metadata such as law_id, paragraph, annex, and chunk.

The workflow requires Qdrant for answer generation when retrieval is enabled. The index job requires the target collection to exist before it runs.