Steps communicate through a scoped variable namespace. A step writes named outputs; later steps reference them in their config with {{varName}} templates, which are resolved at execution time.
Available variables
input.*** — fields from the request body.input._query.*** — query-string parameters.- `input._path.apiname` / `input._path.pathid` — the URL path parameters.
steps.<stepName>.*** — the output of a named earlier step.- `context.userId` — the authenticated user id from the JWT.
- `context.orgCode` / `context.tenantCode` — org and workspace from the request.
- `context.now` — the current ISO timestamp.
Example — RAG over a policy entity
text
s1: query_entity mode=search entity=policy search={{input.question}} limit=3
s2: ai_call intent=generate
contextFields=[
{ label: question, value: {{input.question}} },
{ label: documents, value: {{s1.items}} }
]
instructions="Answer only using the provided documents. If not found, say so."
outputFormat=textTemplates work everywhere
You can use {{...}} in step config values — filter values, URL templates, AI context fields, field mappings. The engine resolves them against the current variable scope before the step runs.
Example — conversation memory
text
s1: list_entities entity=conversation_turn
filters=[{ field: session_id, op: eq, value: {{input.session_id}} }]
sort=created_at sortDir=asc limit=20
s2: ai_call contextFields=[{ label: history, value: {{s1.items}} },
{ label: message, value: {{input.message}} }]
s3: create_entity entity=conversation_turn
fields=[{ session_id: {{input.session_id}} }, { role: user }, { content: {{input.message}} }]
s4: create_entity entity=conversation_turn
fields=[{ session_id: {{input.session_id}} }, { role: assistant }, { content: {{s2.response}} }]