Templating

Most task configuration fields support templates. Before a task processes an event, its configuration is rendered against the event data — so endpoints, query parameters, table names, and other fields can pull values out of the incoming event.

Flowgen uses Handlebars syntax: {{ path.to.value }}.

Where templating runs

For every incoming event, the task’s configuration is cloned and re-rendered with the event as input. This happens before the task’s handler runs, so templates always see the current event:

event arrives → render(config, event) → handler(event)

A task’s static YAML thus becomes a per-event configuration. Connectors that load credentials, schemas, or queries do so against the rendered config.

Available variables

Examples

HTTP endpoint with a path parameter:

- http_request:
    name: lookup
    endpoint: "https://api.example.com/users/{{event.data.user_id}}"
    method: GET

Routing key derived from event metadata:

- nats_jetstream_publisher:
    name: publish
    subject: "orders.{{event.meta.region}}.created"

Pulling a secret from environment:

- http_request:
    name: enrich
    endpoint: "https://api.example.com/data"
    headers:
      Authorization: "Bearer {{env.PARTNER_API_TOKEN}}"

Composing a key prefix from event data:

- nats_kv_store:
    name: cache_user
    operation: put
    key: "users.{{event.data.id}}"
    value: "{{event.data}}"

Type preservation for object and array values

When a template is exactly {{ path }} (no surrounding text, no helpers), flowgen reads the value from the data tree directly instead of converting it to a string. This means objects, arrays, booleans, and numbers retain their original types:

- nats_jetstream_publisher:
    name: forward
    # The whole event.data object is published as-is, not stringified.
    payload:
      from_event: false
      object: "{{event.data}}"

- gcp_bigquery_job:
    name: load
    operation: create
    # source_uris stays a list, not a string representation of one.
    source_uris: "{{event.meta.source_uris}}"

If you need a stringified value, embed the template inside other text: "key.{{event.data.id}}" always renders as a string.

Booleans and numbers in fields that expect them

Numeric and boolean fields (such as port, enabled, max_results) accept template strings that resolve to a number or boolean. The renderer parses the substituted result as JSON and re-typed it before deserialising the config:

- http_request:
    name: paged
    endpoint: "https://api.example.com/items"
    headers:
      X-Page-Size: "{{event.data.page_size}}"

What templating does not do

• It does not run on every field — fields excluded from rendering (binary blobs, sensitive secrets, raw bytes) stay as-is. Most string and structured fields are rendered.
• It does not loop or iterate. Use the iterate task for fan-out over arrays.
• It does not call functions or helpers (no {{#if}}, {{#each}}, etc.). For conditional logic, use a script task to compute the value, then template against the script’s output.
• It does not silently default missing variables — referencing a missing path produces a runtime error.

Templates and resource files

Resource-loaded content (SQL queries, prompts, HTML templates) is also rendered. See Resources for inline-vs-file-vs-cache trade-offs.

- gcp_bigquery_query:
    name: daily_report
    query:
      resource: queries/daily_report.sql
    parameters:
      report_date: "{{event.data.date}}"

The SQL file itself can use the same {{ event.data.x }} syntax — flowgen renders the loaded content against the event before executing the query.