Workflow Nodes

OneHazel is an AI-powered iPaaS (integration platform as a service) that lets you connect external APIs and orchestrate them into workflows — event-driven automations that run without you writing any code.

This documentation covers three concepts:

• Connectors — pre-built integrations for services like Stripe, HubSpot, Postmark, and ~280 others. Connectors define how to talk to a specific API.
• Connections — your authenticated instance of a connector (API keys, OAuth tokens, etc.). You can have multiple connections per connector (e.g. one Stripe connection for test mode and another for live).
• Nodes — the building blocks of a workflow. Each node does one specific thing: call an API, branch on a condition, transform data, delay, retry, etc.

How a workflow runs

1. A trigger node starts the workflow. Triggers can be event-driven (when an event matching a filter is received), schedule-driven (cron), or webhook-driven (external system POSTs to a unique URL).
2. The engine walks the graph, executing each downstream node in order. Branching nodes (If/Else, Switch, Try/Catch, Filter, Retry) decide which outgoing edges fire based on the node's outputs.
3. Each node writes its result into an implicit execution context that downstream nodes can reference using {{node_id.path.to.field}} template syntax.
4. When every leaf branch finishes, the workflow execution is marked complete. You can view every past run under the Workflows tab.

Referencing data between nodes

Any node that accepts a data reference field (marked with the branch icon in the config drawer) lets you pull values from earlier nodes using {{...}} syntax:

{{trigger.data.amount}}
{{stripe_call.data.customer_id}}
{{loop.index}}

The data picker inside those fields renders a tree of every value any upstream node has produced. Pick from the tree and it inserts the correct reference token — no need to type the path manually.

Where to go next

• Triggers — how workflows start (Event, Schedule, Webhook).
• Actions — how workflows do things (API Call, HTTP Request, Emit Event, Execute Workflow, Respond to Webhook, Notify).
• Flow Control — If/Else, Switch, Loop, Parallel, Merge, Delay, Stop.
• Data — Transform, Filter, Aggregate.
• AI — AI Transform, AI Classify, Structured Output, and the AI Agent (LLM-powered steps inside a workflow — including one that can call your connections as tools).
• Error Handling — Try/Catch and Retry for resilience.
• Connectors — reference pages for every connector in the marketplace.

AI nodes at a glance

Node	Use it for
AI Transform	One prompt in, free-form (or JSON) answer out. Summaries, rewrites, extractions.
AI Classify	Pick one of several labels you define; routes the workflow down the matching branch.
Structured Output	Force a JSON object matching a schema you define, validated before it flows on.
AI Agent	Give it a goal + your connections as tools; it decides which to call and loops toward the goal.

Composing workflows

• Execute Workflow lets one workflow call another as a reusable sub-step — factor shared logic out and call it from many places. Pair it with a Return Value node in the sub-workflow to hand a clean result back to the caller.
• Respond to Webhook turns a webhook-triggered workflow into a real request/response endpoint: it returns a status + body to the caller instead of a background 202.