PROCESS NODE

Execute all compiled skills against matched world model nodes. No model calls. No tokens. Pure graph computation.

The LLM's work is already done — it compiled each skill's rule into a graph pattern when the skill was registered. PROCESS NODE runs those patterns directly against your data at graph-native speed.

Syntax#

PROCESS NODE (variable:Label)

Example#

PROCESS NODE (n:Robot)

For each matched Robot node, the engine:

1. Finds all active skills where allowed_on includes Robot
2. Executes each skill's compiled pattern — a WorldCypher graph traversal
3. Scores candidate edges using observation confidence from existing relationships
4. Materializes new edges with confidence score and provenance metadata

No model is consulted. The compiled pattern is the complete specification.

What gets produced#

Every materialized edge carries:

Performance#

PROCESS NODE is a graph traversal operation. Cost scales with the number of matched nodes and the complexity of the compiled pattern — not with edge count or world model size. At 10,000 robots with a co-occupancy pattern: single-digit milliseconds.

There is no network call, no external service, no queue. The pattern runs in the graph engine's process.

Scoped execution#

Run skills against a filtered subset of nodes:

-- Only process robots currently active
PROCESS NODE (n:Robot)
WHERE n.status = 'active'
 
-- Only process robots in a specific zone
MATCH (r:Robot)-[:OCCUPIES]->(z:Zone {id: 'lab-a'})
PROCESS NODE (r)

Multiple active skills#

Multiple skills can target the same label. PROCESS NODE runs all of them:

-- Registered skills:
--   ZonePeerDetector    — co-occupancy relationships
--   CoDetectionLinker   — shared sensor detection relationships
--   CoverageGapFinder   — identify robots without sensor coverage
 
PROCESS NODE (n:Robot)
-- Runs all three compiled patterns
-- Produces ZONE_PEER, CO_DETECTED, and COVERAGE_GAP edges

See Also#

• CREATE SKILL — where the LLM compiles your rule (once)
• REPROCESS EDGES — refresh low-confidence edges, also model-free
• Concepts: Skills — the compile-once model explained