Get Shard Info on Qdrant

Initiates shard movement POST /collections/{collection_name}/cluster/move_shard. Returns async operation status. Use during cluster rebalancing.

Creates a collection alias POST /collections/aliases. Returns status. Use for zero-downtime reindex with QD_SWITCH_ALIAS.

Creates a collection with PUT /collections/{collection_name} and vectors size/distance JSON. Returns acknowledgment. Use QD_CREATE_COLLECTION_AND_WAIT for wait=true.

Universal query API POST .../points/query combining nearest/recommend/discover modes per body. Returns mode-specific results. Use when experimenting; prefer specialized tools for fixed workflows.

GET single point /collections/{collection_name}/points/{id}. Returns point detail. Use for spot checks; bulk reads use QD_GET_POINTS.

Atomically retargets an alias PUT /collections/aliases (action: rename/swap per Qdrant). Supply body JSON per API. Use for zero-downtime reindex cutover.

Deletes a collection via DELETE /collections/{collection_name}. Returns true when removed. Irreversible; export snapshots first with QD_CREATE_SNAPSHOT.

Same as QD_UPSERT_POINTS but always passes wait=true so Qdrant acknowledges persistence before returning. Use when downstream search must see writes immediately.

Batch universal queries POST .../points/query/batch. Returns multiple result sets. Use for throughput; mirrors QD_QUERY semantics.

Returns version, title, and commit from GET /. Use to verify connectivity before heavy operations. Prefer QD_HEALTH_CHECK for load-balancer probes.

Reads anonymized telemetry JSON from GET /telemetry when enabled. Returns usage counters. Use for support bundles; disable telemetry if policy forbids.

Creates collection then polls GET /collections/{name} until status is green or timeout. Returns last describe. Use in automation; interactive flows can use QD_CREATE_COLLECTION only.

Grouped search POST /collections/{collection_name}/points/search/groups to deduplicate by a payload key. Returns group leaders per group_id. Use for catalog dedup RAG; flat lists use QD_SEARCH.

Counts points matching filter POST /collections/{collection_name}/points/count. Returns count. Use before expensive deletes; exact totals may depend on indexing.

Creates a snapshot POST /collections/{collection_name}/snapshots. Returns snapshot name/path. Use for backups; list with QD_LIST_SNAPSHOTS.

Shortcut search: only collection_name, query_vector JSON array, and limit (default 10). Returns hits with payload. Use when no filter is needed; filters belong in QD_SEARCH_WITH_FILTER.

Batch discovery POST /collections/{collection_name}/points/discover/batch for multiple context sets. Returns grouped discovery results. Use for exploratory UIs at scale; single calls use QD_DISCOVER.

Recommends using positive/negative example ids POST .../points/recommend. Returns similar points. Use for “more like this”; vector queries use QD_SEARCH.

Example-first search: supply collection_name, query_vector JSON, and filter JSON (Qdrant Filter object). Returns filtered neighbors. Use for metadata gating; free-form filters still supported via QD_SEARCH body.

Deletes payload keys POST /collections/{collection_name}/points/payload/delete. Returns status. Use to strip fields; for full point removal use QD_DELETE_POINTS.

Runs multiple searches in one POST .../points/search/batch. Returns array aligned to requests. Use for multi-query RAG; single query should use QD_SEARCH.

Lists peers GET /cluster/peers. Returns URIs and ids. Use before QD_REMOVE_PEER maintenance.

Retrieves points by ids POST /collections/{collection_name}/points. Returns vectors and payloads. Use when ids are known; listing unknown ids uses QD_SCROLL_POINTS.

Deletes a payload index DELETE /collections/{collection_name}/index/{field_name}. Returns status. Use to drop unused indexes and save memory.

Deletes points by filter or ids POST /collections/{collection_name}/points/delete. Returns status. Use surgical cleanup; bulk scans may prefer QD_SCROLL_POINTS first.

Cluster metadata GET /cluster. Returns peers and raft info when clustering enabled. Use on self-hosted clusters; single nodes return minimal data.

Lists collections with GET /collections and optional offset/limit pagination. Returns names and basic info. Use before QD_GET_COLLECTION.

Applies multiple upserts/deletes in one POST /collections/{collection_name}/points/batch. Returns per-operation results. Use for throughput; simpler single ops use QD_UPSERT_POINTS.

Vector search POST /collections/{collection_name}/points/search with vector, limit, filter, with_payload, with_vector, params. Returns scored hits. Example filter: {"must":}.

Shortcut: QD_SEARCH with merged body default limit 5 and with_payload true. Returns top five neighbors. Use for tight RAG windows; increase via QD_SEARCH.

Scrolls points POST /collections/{collection_name}/points/scroll with optional filter and offset/limit for full scans. Returns next_page_offset. Paginate until null.

Describes vectors config, points count, and status via GET /collections/{collection_name}. Use before searches; use QD_LIST_COLLECTIONS to discover names.

Exports Prometheus metrics from GET /metrics. Returns plain text exposition format. Use for monitoring; not JSON—parse with metrics tooling.

Shortcut: search with with_vector false to trim bandwidth. Returns ids, scores, payloads. Use in production RAG; enable vectors in QD_SEARCH when debugging embeddings.

Scroll-style browse: POST /collections/{collection_name}/points/scroll with filter only (no vector) via body JSON. Returns points matching payload constraints. Use for admin exports; similarity needs QD_SEARCH.

Patches optimizers, HNSW, or quantization via PATCH /collections/{collection_name}. Returns status. Use for tuning; destructive deletes use QD_DELETE_COLLECTION.

Sets or merges payload fields POST /collections/{collection_name}/points/payload. Returns status. Use for metadata updates without changing vectors.

Discovery search with context pairs POST .../points/discover. Returns exploratory neighbors. Use for exploratory UI; standard kNN remains QD_SEARCH.

Upserts points PUT /collections/{collection_name}/points?wait=true|false with points array. Returns operation status. Use QD_UPSERT_POINTS_AND_WAIT for wait forced true.

Removes a peer POST /cluster/peers/remove with body {peer_id}. Returns status. Destructive—use only during controlled downsizing.

Lists payload indexes GET /collections/{collection_name}/index. Returns indexed fields. Use before QD_CREATE_PAYLOAD_INDEX to avoid duplicates.

Creates payload index PUT /collections/{collection_name}/index for keyword/integer/float/geo/text fields. Returns status. Improves filter speed; unnecessary for tiny collections.

Liveness probe via GET /healthz. Returns ok when process is up. Use for orchestration; pair with QD_GET_TELEMETRY for diagnostics.

Batch recommend POST .../points/recommend/batch. Returns grouped results. Use for catalog personalization at scale.

Lists snapshots GET /collections/{collection_name}/snapshots. Returns snapshot metadata with pagination when available. Use before recovery operations.

Clears all payload POST /collections/{collection_name}/points/payload/clear. Returns status. Use for privacy scrub; vectors remain until deleted.

Lists aliases GET /collections/aliases. Returns mapping of alias to collection. Use before QD_SWITCH_ALIAS.

Deletes an alias via DELETE /collections/aliases/{alias_name}. Returns status. Use after cutover cleanup.

Shortcut search with default limit 50 for recall-heavy retrieval. Returns larger candidate sets. Use before reranking; not ideal for latency-sensitive chat alone.