Querying

ShrikeDB exposes several query surfaces. Use Cypher for graph and multimodal composition, direct vector APIs for ANN search, direct KV APIs for payload access, and SPARQL for supported RDF-style graph pattern queries.

For the full current query guide, see Querying ShrikeDB. That section covers natural-language semantic plans, Cypher, SPARQL, GDS-style analytics, weighted path procedures, path aliases, and result inspection.

Query Surface Map

Need Best surface Why

Need	Best surface	Why
Graph pattern or multimodal retrieval	`ExecuteCypher`	Combines graph patterns with `CALL vector.knn`, `CALL kv.get`, `SEARCH`, and `RANK`.
Vector-only search	`VectorSearch`, `VectorSearchPacked`, `VectorSearchBatchPacked`	Avoids query planning overhead and returns ANN results directly.
KV lookup or batch hydration	`GetKeyValue`, `BulkGetKeyValue`	Direct payload access.
RDF/SPARQL graph pattern	`ExecuteSPARQL`	Useful for SPARQL-shaped graph scans.
Known triple pattern	`GraphScan`	Direct graph scan by subject/predicate/object pattern.

Graph pattern or multimodal retrieval

ExecuteCypher

Combines graph patterns with CALL vector.knn, CALL kv.get, SEARCH, and RANK.

Vector-only search

VectorSearch, VectorSearchPacked, VectorSearchBatchPacked

Avoids query planning overhead and returns ANN results directly.

KV lookup or batch hydration

GetKeyValue, BulkGetKeyValue

Direct payload access.

RDF/SPARQL graph pattern

ExecuteSPARQL

Useful for SPARQL-shaped graph scans.

Known triple pattern

GraphScan

Direct graph scan by subject/predicate/object pattern.

Cypher

Cypher is the primary public graph query entry point.

Implemented syntax includes practical coverage for:

MATCH, OPTIONAL MATCH, WHERE, and RETURN.
WITH, UNWIND, aggregation, DISTINCT, and branch unions.
Null-aware expressions, comprehensions and bounded read-only subqueries.
ORDER BY, SKIP, and LIMIT.
CALL … YIELD procedures.
CALL vector.knn(…).
CALL kv.get(…).
SEARCH collection AS alias.
RANK rerank_v3(…).
Selected optimized count motifs.
Governed CREATE, MERGE, node-property updates and node deletion.
EXPLAIN, PROFILE, and a scoped schema catalog.

See Cypher And openCypher Compatibility for supported shapes and known differences from other Cypher engines.

Example:

CALL vector.knn("products_mm_vec", [0.031, -0.118, 0.442], 50, {ef: 256})
YIELD node AS product, score
MATCH (product:Product)-[:IN_CATEGORY]->(category:Category)
CALL kv.get("default", product) YIELD value AS product_json
RETURN product, category, score, product_json
LIMIT 12

PutKeyValue writes the payload to the default namespace under the product UID. See Multimodal queries for the load and query sequence.

SEARCH

SEARCH is collection-oriented node search sugar. It lowers to a label-constrained scan.

SEARCH products AS product
WHERE product.price >= 50 AND product.stock > 0
RETURN product
LIMIT 12

RANK

RANK rerank_v3 sorts the current row stream by a score binding and can cap candidates before projection or final LIMIT.

CALL vector.knn("products_mm_vec", [0.1, 0.2], 100) YIELD node, score
RANK rerank_v3(top_k = 50, score = score)
RETURN node, score
LIMIT 12

Supported options:

top_k: positive integer.
score: variable or string literal naming the score binding.

MERGE And SET

With governed writes enabled and a write-authorized query surface, a supported node-upsert shape is:

MERGE (a:Account {id: 'acct_a', run_id: 'run_1'})
SET a.last_seen = 42, a.active = true
RETURN a
LIMIT 1

The general planner lowers node and directed single-hop merges, including supported ON CREATE SET and ON MATCH SET actions, to governed write operators. The default database also retains a legacy MERGE … SET … RETURN fast path that does not consult the governed write-mode setting. See Cypher Writes And Schema for enablement, deterministic-match requirements, creation limits and unsupported mutation forms.

Planned Concept Syntax

The following concept syntax appears in examples and roadmap docs but is not current parser syntax:

EMBED("vintage leather jacket") AS q
FULLTEXT MATCH "leather"

Use client-side embeddings, pre-fused vector indexes, or explicit CALL vector.knn today.

SPARQL

Use ExecuteSPARQL for supported graph-pattern queries. The distributed HA read path explicitly covers graph-scan-shaped SPARQL such as:

SELECT ?s ?o
WHERE {
  ?s <http://example.com/ns/edge> ?o .
}

See SPARQL for additional read-query forms and restrictions.

Graph Scan

Use GraphScan when you already know the triple pattern.

subject:   "?person" or "uid:123"
predicate: "?p" or "knows"
object:    "?friend" or "uid:456"

Direct Vector Search

Use packed responses when the caller only needs IDs and scores.

VectorSearchPacked(index_name, query_vector, k, ef_search)
  -> entity_ids[]
  -> scores[]

Use batch packed search to send multiple query vectors in one RPC.