qdrant

Vector embeddings enable semantic search by storing numerical representations of text or content in a vector database. Qdrant stores these embeddings in collections and retrieves similar vectors using distance metrics like cosine, dot product, or Euclidean distance, making it ideal for finding semantically related documents without keyword matching.

Yes, Qdrant is designed for RAG workflows. It stores document embeddings with metadata payloads, then retrieves contextually relevant documents during generation. You upsert vectors with associated data, query for similar embeddings, and optionally filter by payload conditions to feed retrieved context into language models.

Upsert vectors into Qdrant collections via REST API using curl. Each vector includes an ID, embedding values, and optional payload metadata. The payload stores contextual information like document text or tags, allowing you to attach and filter on domain-specific attributes alongside similarity search.

Qdrant supports cosine distance, dot product, and Euclidean distance for measuring vector similarity. Choose based on your embedding model and use case: cosine is common for normalized embeddings, dot product for pre-normalized vectors, and Euclidean for geometric distance. Distance type is set at collection creation.

Yes, this Skill uses Qdrant's REST API accessed via curl commands. You interact with collections and points by constructing HTTP requests to manage vectors, perform searches, and apply filters. Basic familiarity with curl and JSON payloads is required; authentication uses environment variables QDRANT_URL and QDRANT_API_KEY.

Yes, Qdrant supports filtering search results by payload conditions. Combine vector similarity with payload filters to narrow results—for example, searching for similar embeddings only within a specific document category or date range, combining semantic relevance with structured metadata constraints.