Triggers

A trigger is a pre-planned response to a graph event. You declare it before the event happens. When the event occurs — a node is created, a property changes, a node is deleted — the trigger fires once and does its work.

-- When a frame arrives, run the detection skill
CREATE TRIGGER detect_on_frame
    ON :ImageFrame WHEN CREATED
    RUN SKILL detect_objects

This is different from a live query. A live query maintains a result set continuously — it is always true. A trigger fires once per matching event — it is a deliberate response.

Choose LIVE when you need a result set that is always current. Choose TRIGGER when you need to act once in response to a specific event.

Syntax#

CREATE TRIGGER#

CREATE TRIGGER <name>
    ON :<Label> WHEN CREATED | MODIFIED | DELETED
    RUN SKILL <skill_name>

ON :<Label> — The label that determines which nodes are watched. Only nodes with this label will match.

WHEN CREATED — Fires when a new node with the label enters the graph.

WHEN MODIFIED — Fires when a property on a matching node changes.

WHEN DELETED — Fires when a matching node is removed from the graph.

RUN SKILL <name> — The skill to execute. The skill receives the triggering node as input.

DROP TRIGGER#

DROP TRIGGER detect_on_frame

Deregisters the trigger. In-flight executions are not interrupted.

Event types#

CREATED — respond when a node arrives#

-- Run enrichment whenever a new Person node is added
CREATE TRIGGER enrich_person
    ON :Person WHEN CREATED
    RUN SKILL extract_relationships
 
-- Start detection when a new frame is captured
CREATE TRIGGER detect_on_frame
    ON :ImageFrame WHEN CREATED
    RUN SKILL detect_objects
 
-- Index a document when it enters the workspace
CREATE TRIGGER index_document
    ON :Document WHEN CREATED
    RUN SKILL embed_and_link

MODIFIED — respond when state changes#

-- Re-score when a detection's confidence is updated
CREATE TRIGGER rescore_on_update
    ON :Detection WHEN MODIFIED
    RUN SKILL score_and_link
 
-- Recalculate risk when an entity's position changes
CREATE TRIGGER position_risk_update
    ON :Entity WHEN MODIFIED
    RUN SKILL recalculate_risk_score

DELETED — respond when a node leaves#

-- Clean up linked evidence when a session ends
CREATE TRIGGER cleanup_on_session_end
    ON :Session WHEN DELETED
    RUN SKILL archive_session_evidence

Trigger vs Live: choosing the right primitive#

The same scenario can often be written two ways. The semantics are different.

Scenario: index documents as they arrive.

With a trigger — fire once per document:

CREATE TRIGGER index_on_arrival
    ON :Document WHEN CREATED
    RUN SKILL embed_and_link

With a live query — maintain the set of unindexed documents continuously:

CREATE LIVE VIEW unindexed_documents AS
  MATCH (d:Document)
  WHERE d.indexed IS NULL OR d.indexed = false
  RETURN d.id, d.content, d.created_at

The trigger is better when the action is a one-time side effect — index this document, enrich this node, start this workflow. The live query is better when you need to know the current state — what documents still need indexing — across restarts, failures, or batches.

Use a trigger when:

• The action is a side effect that should happen once per event
• You are invoking a skill against each new or changed node
• You are wiring a program's output stage to a downstream skill

Use a live query when:

• You need a continuously maintained result set
• Multiple consumers read the same derived data
• The view should survive session ends and restarts

Triggers inside Programs#

Programs can declare triggers inline. This is the recommended pattern for production deployments — the trigger's lifecycle is tied to the program's install/remove cycle:

CREATE PROGRAM yolo_v11 VERSION '1.0' (
    ...
    SKILLS [detect_objects, score_balls],
    TRIGGERS [ON :ImageFrame WHEN CREATED]
)

When the program is installed, the trigger is registered. When the program is removed with DROP PROGRAM, the trigger is dropped. There is no orphaned trigger to clean up.

Standalone CREATE TRIGGER is for cases where the trigger is not part of a program — ad hoc automation, scripted pipelines, or development workflows.

Inspecting triggers#

-- List all registered triggers
CALL db.triggers()
YIELD name, label, event, skill, created_at
RETURN *

| name                   | label      | event   | skill            | created_at  |
|------------------------|------------|---------|------------------|-------------|
| detect_on_frame        | ImageFrame | CREATED | detect_objects   | 1744600000  |
| enrich_person          | Person     | CREATED | extract_rels     | 1744600100  |
| rescore_on_update      | Detection  | MODIFIED | score_and_link  | 1744600200  |

See also#

• Programs — CREATE PROGRAM bundles triggers, skills, and executor endpoints into a single installable manifest
• Live Queries — CREATE LIVE VIEW and LIVE MATCH for continuously maintained result sets
• Skills — CREATE SKILL, PROCESS NODE, REPROCESS EDGES
• Event Sourcing — CDC and mutation log for downstream event pipelines