Get similar-item recommendations from Qdrant using the recommend query type with positive/negative examples

Documented steps

1. Auth: api-key header if API-key auth is enabled (always required on Qdrant Cloud).
2. Use the Query API with a nested recommend query: POST /collections/{collection_name}/points/query with {"query": {"recommend": {"positive": [100, 231], "negative": [718, [0.2,0.3,0.4,0.5]]}}, "limit": 3} — mixing point IDs and raw vectors is allowed
3. Avoid the older POST /points/recommend endpoint for new code — deprecated in favor of the /points/query recommend query type
4. Add a filter alongside the recommend query to restrict candidates, e.g. {"key": "city", "match": {"value": "London"}}
5. Choose a strategy: average_vector (default; avg(positive) + (avg(positive) - avg(negative)); requires ≥1 positive) or best_score (v1.6.0+; scores against each example individually, works with only negatives, more diverse results)
6. For multi-vector collections set "using": "<vector_name>" so recommendation runs on the intended vector space
7. curl: curl -X POST 'http://localhost:6333/collections/{collection_name}/points/query' -H 'api-key: <API_KEY>' -H 'Content-Type: application/json' -d '{"query":{"recommend":{"positive":[100,231],"negative":[718]}},"strategy":"best_score","limit":10}'
8. Official docs: https://qdrant.tech/documentation/concepts/explore/ ; https://api.qdrant.tech/api-reference/search/query-points