Workflows appear only for workspaces where the feature is enabled. Available steps and integration actions depend on the integrations connected to your workspace.
Workflow list
Open Workflows in the Dashboard to view existing workflows. From the list, you can:- Search by workflow name
- Filter by Draft, Published, or Disabled
- Create a workflow
- Edit a workflow
- View workflow runs
- Duplicate a workflow
- Delete a workflow, which archives it and removes it from the list
Create a workflow
- Open Workflows.
- Select Add workflow.
- Enter a name.
- Add one or more triggers to the Start node.
- Add steps from the step panel and connect them in the order they should run.
- Configure each step’s inputs, conditions, mocks, and outputs as needed.
- Test the workflow with mock input.
- Publish the workflow when it is ready to run from its configured triggers.
Workflow statuses
In the editor, open More actions to find Pause runs, Resume runs, Duplicate workflow, and Archive workflow. Use Pause runs to stop triggers from firing for a published workflow. Use Resume runs to allow triggers to fire again. Draft and archived workflows cannot be paused or resumed.
If another workflow uses this workflow as a sub-workflow, remove it from that workflow before you delete or archive it.
Triggers
Triggers define when a workflow starts. Configure triggers from the Start node.
Each trigger can include:
- A name
- Optional conditions that decide whether the workflow should run
- Input mapping that transforms trigger data into the workflow input
- Trigger-specific configuration, such as the selected event or the AI trigger’s settings
- Instructions tell the AI when to run the workflow and what to put in the input.
- Input shape defines the fields the AI supplies. Map them into the workflow input with
data.input.
conversationId and conversationExternalId, the conversation’s ID in your helpdesk after it syncs there. When the AI calls the workflow in Playground, Remark skips the run.
HTTP triggers use the endpoint shown in the trigger panel:
data in the trigger’s input mapping.
When the request starts a run, the response includes the run’s id, status, and created time. If the trigger’s conditions don’t match, the endpoint returns 200 with {"skipped": true, "reason": "..."} and creates no run.
Steps
Steps are the work a workflow performs after it starts. Add steps from the Steps panel, then connect them on the canvas. Available steps can include:
Some workspaces can also have browser-session steps for conversation-specific workflows. Use those only when the workflow is explicitly designed around an active shopper conversation.
Add notes with Add note in the canvas toolbar. Notes never run.
A workflow’s output is the output of its final step. When a workflow ends in more than one step, its output is a list with a
{ stepKey, output } entry for each final step that ran. A parent workflow reads a sub-workflow’s output from the sub-workflow step’s output.
Step configuration
Most runnable steps have the same configuration concepts:
The Start node can also define the workflow input shape and mock input. Use these fields to make tests realistic before the workflow is published.
Test and run
Use Test to run the latest draft with the Start node’s mock input. Most steps return their mock output instead of running, so a test has no real side effects. Transform and Evaluate steps, and HTTP steps that sendGET requests, run for real. Test runs are useful for checking input mapping, step order, conditions, and expected outputs.
Use Run to run the published workflow with real side effects. A workflow must be published before it can run. Test and Run share a button that repeats the mode you used last. Open its menu to switch modes.
Publish, Test, and Run stay blocked until you fix the errors highlighted on the canvas, such as a step missing a required input.
Runs
Open the Runs tab to review workflow executions. Runs can have these statuses:
The run detail view shows the trigger type, trigger data, workflow input, workflow output, duration, errors, and per-step results. Select a step in a run to inspect its input, output, status, duration, error message, and retry attempts when available.
Each run shows how it started: AI, Event, HTTP, Manual (the Run control), Test, or Workflow (a parent workflow). Steps can be Pending, Running, Waiting, Completed, Failed, Skipped, or Cancelled.
Select Stop run to cancel a pending or running run. Select Run again to start a new run with the same input. A test run repeats as a test of the latest draft. Any other run repeats as a real run of the published workflow, with real side effects.
Troubleshooting
A workflow did not run
Check that the workflow is published, not disabled, and has at least one trigger configured. If the trigger has conditions, confirm the trigger data satisfies those conditions.An HTTP trigger returns an error
Confirm the request usesPOST, sends Authorization: Bearer <VendorApiKey>, and uses a vendor API key with Triggers workflows permission. The workflow must be published, not paused, and have exactly one HTTP trigger. If the body doesn’t match the workflow’s input shape, the endpoint returns 400 with the validation message.
A run failed
Open the run detail and inspect the failed step. Check the step input, output, error message, and any trigger data shown on the Start node. For integration actions, also confirm the connected integration is healthy and the action supports workflows.A step did not run
If a step is skipped, review its condition and the outputs from earlier steps. Notes are canvas-only and never appear as executed steps.Related pages
- Use Intents when the AI needs a customer-facing playbook for a shopper scenario.
- Use Actions to understand the single-step capabilities that intents and workflows can call.
- Use Inbox to review the conversations that can trigger AI or event-driven workflows.
- Use Playground to test AI behavior before relying on it in live conversations.
- Use Custom Integrations when a workflow depends on external commerce, support, sizing, or order data.