Querying ShrikeDB

ShrikeDB has more than one query surface because different tasks need different interfaces. Choose an interface based on your data model and whether you need direct query control or a profile-guided question.

Use Cypher for property-graph patterns and multimodal composition, SPARQL for RDF and result graphs, vector RPCs for direct nearest-neighbor retrieval, SpaceKey for bounded space-time observations, and semantic MCP for profile-guided analysis.

Choose The Right Surface

You want to Use Why

Ask a business or investigation question

Natural language / semantic query

ShrikeDB uses the selected profile to produce a typed semantic plan, explain it, and execute it with evidence.

Run graph patterns, analytics, or GDS-style procedures

Cypher

Graph patterns, optional matches, aggregation, pipelines, bounded subqueries, paths, vector/KV composition and plan inspection; governed writes use a separate enablement boundary.

Query RDF-like triples or ontology facts

SPARQL

Best direct surface for SELECT graph patterns and ShrikeDB SERVICE extensions.

Let an agent reason through prompts and typed intents

Semantic MCP

The agent receives XML workflows, returns structured intent, and never has to invent raw database queries.

Debug a precise graph, vector, or KV primitive

Direct gRPC or admin/API tools

Useful for loaders, tests, and low-level health checks.

Start with Languages and interfaces for a complete map of the public surfaces. The dedicated vector chapter covers named indexes, insertion, streaming and packed search, Cypher composition, tuning, and recovery. See Multimodal and hybrid queries when one request combines graph, vector, KV, RDF, SpaceKey, or lakehouse data.

What The Frontend Shows

The chat UI separates the answer from the evidence:

  • Data is the row set returned for the answer.

  • Visuals are generated from the same rows when a table, graph, or chart makes sense.

  • Reasoning explains the selected semantic family and plan.

  • Knowledge shows profile, ontology, lakehouse, and evidence metadata.

  • Query shows the typed plan and, for natural-language answers, a Cypher-equivalent inspection view.

  • Inspect shows lower-level traces, result hashes, substrate choices, and diagnostics.

Natural-language answers usually execute typed plans, not literal Cypher. Read a Cypher-equivalent plan as an explanation of the typed plan; direct Cypher mode submits literal query text.

Query Safety Rules

ShrikeDB prefers a clear rejection over a convenient wrong answer. Expect explicit diagnostics for:

  • unsupported unbounded paths;

  • missing profile or ontology context;

  • unsupported weight expressions;

  • unsafe graph analytics write modes;

  • lakehouse mappings that do not cover the requested semantic concepts;

  • direct query text submitted through semantic/chat mode.

Useful Next Pages