varpulis

Temporal pattern detection for n8n — powered by Varpulis CEP engine (WASM)

Package Information

Released: 6 mo ago
Downloads: 4 weekly / 26 monthly
Latest Version: 0.1.0

Documentation

n8n-nodes-varpulis

Add temporal pattern detection to any n8n workflow.

This community node brings the Varpulis CEP engine into n8n. It runs entirely inside your n8n instance via WebAssembly -- no sidecar process, no external service, no network calls. Define patterns in Varpulis Pattern Language (VPL), feed events from any n8n trigger, and route detected matches to downstream nodes.

What it does

The Varpulis node watches a stream of incoming n8n items and detects temporal patterns across them -- sequences of events that occur in a specific order within a time window, with conditions on their fields. When a pattern matches, it emits a new item on the Matches output. Every input item is also forwarded unchanged on the Passthrough output.

Use cases:

  • Payment fraud: detect rapid sequences of failed-then-succeeded transactions
  • IoT anomaly: catch temperature spikes followed by pressure drops within a window
  • User behavior: identify sign-up-then-churn patterns for engagement workflows
  • Monitoring: correlate error bursts across services in real time

How it works

Trigger --> [Varpulis CEP] --Matches--> Slack / DB / HTTP / ...
                |
                +--Passthrough--> next node
  1. You write a VPL pattern that declares event types and stream definitions.
  2. For each execution, a fresh WASM engine loads the pattern and processes every input item as an event.
  3. Items that trigger a pattern match appear on the Matches output with emitted fields.
  4. All original items pass through unchanged on the Passthrough output.

The WASM engine is ~2 MB and initializes in milliseconds. There is no state between executions.

Installation

In n8n (community nodes)

  1. Go to Settings > Community Nodes
  2. Enter n8n-nodes-varpulis
  3. Click Install

Manual (self-hosted)

cd ~/.n8n/nodes
npm install n8n-nodes-varpulis
# Restart n8n

From source (development)

cd integrations/n8n-nodes-varpulis
npm install
npm run build
# Then link into your n8n instance:
npm link
cd ~/.n8n/nodes
npm link n8n-nodes-varpulis

Node properties

Property Description
VPL Pattern The full VPL program -- event declarations and stream definitions
Event Type The event type name to assign to each incoming item (must match a declared event)

Example: Stripe payment fraud detection

Detect a failed payment followed by a successful payment from the same customer within 5 minutes:

VPL Pattern:

event Payment:
    customer_id: string
    status: string
    amount: float

stream SuspiciousRetry =
    Payment as failed -> Payment as succeeded
    .where(failed.status == "failed"
       AND succeeded.status == "succeeded"
       AND succeeded.customer_id == failed.customer_id
       AND succeeded.amount >= failed.amount)
    .within(5m)
    .emit(
        customer: failed.customer_id,
        failed_amount: failed.amount,
        succeeded_amount: succeeded.amount
    )

Event Type: Payment

Wire a Stripe webhook trigger into the Varpulis node, and connect the Matches output to a Slack notification or database insert.

Example: IoT temperature anomaly

Detect a high-temperature reading followed by a pressure drop within 2 minutes:

VPL Pattern:

event Sensor:
    zone: string
    temperature: float
    pressure: float

stream TempPressureAnomaly =
    Sensor as hot -> Sensor as drop
    .where(hot.temperature > 80
       AND drop.pressure < 30
       AND drop.zone == hot.zone)
    .within(2m)
    .emit(
        zone: hot.zone,
        peak_temp: hot.temperature,
        low_pressure: drop.pressure
    )

Event Type: Sensor

VPL syntax basics

# Declare an event type with typed fields
event <Name>:
    <field>: <type>    # string, int, float, bool

# Define a pattern stream
stream <Name> =
    <EventType> as <alias> -> <EventType> as <alias>  # sequence
    .where(<conditions>)                                # filter
    .within(<duration>)                                 # time window (s/m/h)
    .emit(<field>: <expr>, ...)                         # output fields

Supported types: string, int, float, bool

Duration units: s (seconds), m (minutes), h (hours)

Conditions support: AND, OR, NOT, comparisons (==, !=, >, <, >=, <=), arithmetic (+, -, *, /).

For full syntax including Kleene patterns, negation, aggregations, and forecasting, see the VPL documentation.

Outputs

Output Description
Matches Items emitted when a pattern matches. Contains the .emit() fields plus _stream (stream name), _timestamp, and _sourceIndex (which input item triggered the match).
Passthrough Every input item, unchanged. Use this to continue processing regardless of matches.

Links

License

MIT

Discussion