Skip to main content
Use this reference to pick the right node for each step of your automation. Grouping mirrors the builder palette. Each entry highlights when to use it, key parameters, and outputs.

​
System

  • Start (START) — required entry; one per flow. No params or outputs.
  • End (END) — terminates the run. Add error transitions on upstream nodes for resilience.
  • Navigate (NAVIGATE) — load a URL (optionally new tab). Writes ${node_id}.final_url for downstream use.
  • Click (CLICK) — activate a button/link/element. Add Wait For after view-changing clicks; XPath override supported.
  • Input (INPUT) — type into a field. Accepts ${var} text; XPath override for brittle fields.
  • Select Option (SELECT_OPTION) — choose dropdown option by visible text. Pair with Wait For if the page refreshes.
  • Scroll (SCROLL) — move the viewport up/down by pixels. Chain for long lists.
  • Key Press (KEY_PRESS) — send keys (Enter/Escape/arrows) to advance dialogs or forms.
  • AI Action (BROWSER_ACTION) — one-shot prompt-driven action. Non-destructive; follow with Wait For / Extract Data to verify results.

​
Waiting & readiness

  • Wait (WAIT) — fixed delay (ms) for predictable pauses.
  • Wait For (WAIT_FOR) — block until text/selector appears; use after navigation or interactions that change state.

​
Data capture & transform

  • Extract Data (EXTRACT_DATA) — structured scrape. Define fields (scalar/list); capture_scope sets visible vs full-page.
  • Process Data (PROCESS_DATA) — prompt-based transform. Declare outputs (name, description, type: scalar | list | object | object_list); ${var} tokens interpolated. Add schema fields for object/object_list so downstream steps and API callers see the shape.
  • Parse Document (PARSE_DOCUMENT) — parse a file URL or canonical variable; emits ${node_id}.parsed_result_json and ${node_id}.parsed_result_url.
  • File Output (FILE_OUTPUT) — write text/JSON to storage; exposes path at output_variable_name.

​
Files & downloads

  • Browser Upload (BROWSER_UPLOAD) — attach a file inside the page. Target control; map file variable; XPath optional.
  • Browser Download (BROWSER_DOWNLOAD) — click a download control and wait for Chrome’s download event; stores saved file path for downstream nodes.

​
Communication & access

  • Login (LOGIN) — apply stored credentials at the right moment; keeps secrets out of nodes.
  • Send Email (EMAIL_SEND) — send via Resend; to, subject, and html_body accept ${variable} interpolation.

​
APIs & integrations

  • API Request (API) — call any HTTP endpoint (GET/POST/PUT/PATCH/DELETE) from within a task. Provide URL, optional Bearer token, JSON body with ${variable} interpolation, and a timeout. The JSON response is parsed and stored at ${node_id.output_variable_name} for downstream use. Use the in-builder Send test button to try sample requests safely.

​
Forms & composition

  • Fill Form (FILL_FORM) — set many fields in one node, then verify.
  • Task (TASK) — embed another published task as a child step. Maps parent inputs to child, lets you choose child outputs to surface, and runs the latest published version by default (auto-updates). Switch to “Pin to version” when you need a fixed child definition.

​
Logic & control

  • If / Else (IF_ELSE) — branch on plain-language condition; routes to true/false handles.
  • Loop (LOOP) — iterate over a canonical list variable (no ${}); inside the loop use ${loop_id}.current_item and ${loop_id}.current_item_index.

​
Tips for reliable runs

  • Add Wait For after any action that changes the view before extracting data.
  • Prefer absolute URLs for downloads; use Process Data to prepend host/scheme to relative links.
  • Keep node descriptions short and action-oriented so collaborators quickly understand intent.
  • For AI-driven actions (Browser Action, Fill Form, Login), add Visual references in Advanced options so the agent sees what “success” looks like. Use tight, well-labeled screenshots for best results.