Skip to main content
Advanced workflows (visual workflows) let operators compose deterministic automation graphs on a canvas. They complement agent automations, which use an LLM orchestrator and tool plan. The feature is in Beta. It is gated by the workspace-visual-workflows flag for your organization.

​
Where to find them

Open organization Automations. When advanced workflows are enabled, click Advanced workflows to open the list at /org/{slug}/automations/visual-workflows. From there you can create a workflow, open the canvas editor, and inspect run history per workflow. Advanced workflows are organization-scoped today. They do not appear on project-scoped automation pages.

​
Agent automations vs advanced workflows

Use agent automations when the agent should decide which tools to call. Use advanced workflows when the steps, order, and branches are fixed.

​
Build a workflow

  1. Click New workflow on the advanced workflows list.
  2. Add nodes from the picker: Manual trigger, HTTP request, If / else, or AI agent.
  3. Connect nodes on the canvas. The graph must have exactly one trigger and every non-trigger node must be reachable from it.
  4. Configure each node in the side panel.
  5. Save. Workflows start in draft status; set active when you are ready to run them.

​
Node types (Phase 1)

For each loop nodes appear in the picker but are not executable in Phase 1.

​
Template expressions

HTTP URLs, if conditions, and AI prompts support {{ ... }} placeholders that resolve at run time:
  • {{ trigger.field }} — value from the manual run input
  • {{ nodes.<nodeId>.field }} — output from an upstream node (for example {{ nodes.http-1.body }})
If conditions support comparison operators (===, !==, ==, !=, >, >=, <, <=) after templates are resolved. Empty conditions evaluate to false.

​
Graph validation

Saving rejects invalid graphs:
  • Missing or multiple triggers
  • Edges pointing at missing nodes
  • Nodes not reachable from the trigger

​
Run a workflow

From the editor, queue a manual run. Cloud creates a run record, enqueues execution, and updates each node as it progresses. Run statuses: queued, running, succeeded, failed, cancelled, or skipped. Each run stores a definition snapshot so you can inspect what executed even if the workflow definition changed later.

​
Idempotency

Manual runs require an idempotencyKey. Reusing the same key for the same workflow returns the existing run instead of creating a duplicate.

​
HTTP action constraints

HTTP nodes use the same SSRF-safe fetch path as other public outbound requests in Cloud:
  • Only public HTTPS URLs are allowed
  • Redirects are not followed automatically
  • Non-2xx responses fail the node
Do not put secrets in URLs. Prefer short-lived tokens in trigger input when you must pass credentials.

​
Permissions and API

Advanced workflow routes require an organization operator role (admin or localization manager). The feature flag must also be enabled. Authenticated JSON API (session cookie or equivalent app auth): Responses use resource-keyed envelopes (visualWorkflow, visualWorkflows, run, runs). Errors follow the standard { error, message, details? } shape. When the flag is off, routes return forbidden with visual_workflows_feature_disabled.

​
Troubleshooting

​
Next