Events
Events are the data records that flow between tasks. Each task receives events, processes them, and emits new events to the next task.
Structure
| Field | Type | Description |
|---|---|---|
data | EventData | Payload — JSON, Arrow RecordBatch, or Avro. |
subject | string | Task name that produced the event. |
id | string | Optional identifier set by the source (e.g., Salesforce record ID, NATS message ID). |
timestamp | int | Creation time in microseconds since Unix epoch. |
task_id | int | Task index in the flow. |
task_type | string | Task type (e.g., script, http_request). |
meta | map | Key-value metadata that travels with the event. |
error | string | Error message if a task failed after retries. |
Data formats
- JSON — default for REST APIs, webhooks, scripts. Accessed as
event.data.fieldin templates and scripts. - Arrow RecordBatch — columnar format for BigQuery, Parquet, CSV. Stays in Arrow through the flow without serialization overhead.
- Avro — binary format for Salesforce Pub/Sub and gRPC streams.
Metadata
event.meta is a key-value map that travels through the entire event chain. Each task preserves meta from the upstream event automatically.
Use meta for:
- Passing context between tasks (e.g., source file path, request headers).
- Correlation —
event.meta.correlation_idis auto-generated (UUIDv7) for every event chain. - User context —
event.meta.authcontains resolved user identity for authenticated webhooks.
Set meta in scripts:
event.meta.source = "salesforce";
event.meta.batch_id = uuid();
eventAccess meta in templates:
endpoint: "https://api.example.com/{{event.meta.region}}/orders"Correlation ID
Every event chain gets a correlation_id in meta, generated automatically when the first event is created. It stays the same through all downstream tasks. Use it to trace a single request through the entire flow in logs.
Webhook, AI gateway, and MCP tasks set their own correlation_id for response streaming. These override the auto-generated one.
Event chain
Tasks are connected by channels. Each task receives from the previous task and sends to the next:
[source] → channel → [transform] → channel → [sink]If a task returns no data (e.g., a filter script returns ()), the event chain stops — no downstream event is emitted.