Workflow AI

Workflow AI turns a multi-step job — read these documents, extract the figures, store them, email a summary — into a workflow: a set of connected nodes on a canvas that you can run on demand, on a schedule, from a webhook, or publish as an app. You can draw workflows by hand or describe them and let the AI Assistant build them.

The workflows page

The module opens on your list of workflows.

• Search, filter and sort — by name, tags, or status; sort by last run, last modified or creation date.
• Badges on each card: Appified (published as an app), Scheduled, Shared (shared with you, with edit and run or view only), and the status of the last run (Idle, Running, Completed, Failed, Stopped).
• Card actions — open, Share, Duplicate, Delete (moves to the trash — see Trash).

Create workflow opens the creation hero: "What would you like to automate?" Describe it in plain words and press Generate with AI, or start from a template — Empty workflow, Document summarizer, Web research report, Data extraction to database. Templates are just starting prompts; you can change anything afterwards.

The canvas

A workflow always begins at the Start node. Everything else is added from the node palette on the left, which groups the available nodes by category (document processing, data, web, communication, agents, decisions, …); type in Filter the nodes… to search. Drag a node onto the canvas and connect it by dragging from an output handle to the input of the next node.

Some node kinds are worth knowing by name:

Node	What it does
Start	Entry point. Turn on Enable Webhook Trigger to start the workflow from outside (see Webhooks)
User Input	Asks for text, files or choices when the workflow runs — the form fields of an app
Super Agent	The Chat AI assistant as a node: give it a task in plain words and it plans, uses tools, and returns a result (with an Agent Chat tab to talk to it mid-run)
Document nodes	Read, summarise, extract from or search the files you attach from My Data
Database / DB filter	Read from and write to your table databases and connected SQL sources
Decision	Routes the run down different branches depending on rules you set
Analytics dashboard	Turns data into a chart and pins it to a Dashboard
Webhook	Calls an external URL with data from earlier nodes
Note	A sticky note for documentation; it does not run

Configuring a node

Click a node to open its toolbar on the right. It has up to five tabs:

• Input — the node's parameters. Type a value, or reference the output of an earlier node: start typing {{ and pick from the list, or drag a field from another node's output. Fields with problems are flagged (Missing fields, Invalid fields) and listed in the Issue navigator so you can jump between them.
• Output — the result of the last run, Pretty or Raw, with Copy response and Export Output. Turn on Export Output to make the node's result visible in Apps and in the workflow Outputs view.
• Table — for nodes that produce tables, the rows they wrote.
• Chat — ask questions about the node's output.
• Agent Chat — for Super Agent nodes, the conversation with the agent.

Each node also has a Run button to run that node on its own (use it to test a step before running everything; it needs the outputs of the nodes before it, so run those first) and Stop node. Make editable in app marks a field as something app users can change (see Appify).

You cannot connect a node to itself, connect anything into the Start node, or create the same connection twice; the canvas tells you why.

Saving

Press Save Workflow (or Ctrl/Cmd+S). Unsaved changes shows in the header until you do. A first save creates the workflow; later saves add a version (see History & versions).

If a colleague saved the same workflow while you were editing, saving shows a Workflow save conflict dialog: Use server version discards your canvas in favour of theirs, Fork as new workflow keeps your canvas as a new workflow and reloads theirs here.

Saving while a run is in progress is allowed, but the run may still be using the previous version.

Building with the AI Assistant

The AI Assistant sidebar (right) builds and edits workflows from a description: "Describe what you want to automate — we'll build or update the workflow for you."

• On an empty canvas it creates the whole workflow (Building your workflow…, Filling in the details…). If something is unclear it answers I need more details and asks.
• On an existing workflow it makes changes — "add an email step at the end", "make the summary shorter". Before applying, a review lists what would be added, changed and REMOVED so nothing disappears unseen.
• Select nodes on the canvas first to scope a request to them.
• Assistant replies can include directives — small actions such as Schedule update, Attach file or Previous version restored — save to keep it — with an Applied state and Retry if one fails.
• Stop generating interrupts; Clear chat starts the conversation over.

Assistant changes land on the canvas unsaved: review them, then Save.

Skills

Skills are reusable instruction sets — a procedure, the tools it may use and the output style — that Super Agent nodes and Chat AI can follow. Manage them from the assistant's Skills panel: create one, edit its instructions, Export it as a package, or Import a package shared with you. Attach a skill to a Super Agent node in its Input tab.

Running

• Run workflow runs the whole thing from Start. Nodes light up as they progress: Idle → Running → Completed / Failed. Hover a failed node to read the error.
• Stop workflow cancels a run; the status becomes Stopped.
• Run on a single node re-runs the workflow from that node onward, using the saved outputs of everything before it — handy while iterating on the last steps.
• A workflow with documents attached needs them trained first; the app asks you to wait if an upload or training is still in progress.

Runs count against your plan's monthly quota — see Plans, billing & usage.

Outputs

The Outputs view collects the results of every node that has Export Output enabled, run by run. Filter to All Outputs or the Latest N Outputs, sort by kind or status, copy or download a result. Charts built by the workflow are listed under Linked charts, with Open Charts taking you to the dashboard.

History & versions

The History panel lists every saved version — who saved it and when, with the current one marked Current version. Open an older version to view it (Viewing a previous version. Restore to edit.) and Restore it; the restore is saved as a new version tagged (restored), so nothing is lost.

Schedule

Schedule runs a workflow automatically. Choose a Schedule Type:

Type	Example
Interval	Every 30 minutes
Daily / Weekly / Monthly	Every weekday at 08:00; the 1st of each month
Run once at	A specific date and time
Advanced scheduling (cron)	0 6 * * 1-5

Times are shown in your local timezone. A scheduled workflow shows the Scheduled badge; the schedule can be enabled, disabled or deleted without touching the workflow. Only the workflow's owner can schedule it.

Webhooks

To start a workflow from another system, open the Start node and turn on Enable Webhook Trigger, then save (Save workflow to get the webhook ID). The toolbar shows:

• the Webhook URL to POST to, and
• an example Webhook Payload with your workflow's credentials (webhook_id, webhook_secret) and a webhook_payload object. Any keys you send in webhook_payload become outputs of the Start node that the next node can link to.

Individual nodes can have their own webhook that triggers from this node instead of the whole workflow.

Webhook runs are queued and executed one at a time per workflow. The Run queue (View queue) shows what is Queued, Running, Run completed, Run failed or Cancelled, including files the call added to your data (n file added, n file already in your corpus). You can cancel a waiting run or retry a failed one.

The request format, limits and status polling are in the API reference: Webhooks.

Sharing

Share Workflow offers two very different things:

• Work on it together — no copy is made; the people you add open and edit this workflow, and everyone sees each other's saved changes. Give each person Edit and run or View only. Only you can delete, share, schedule or add other people. A shared workflow shows Shared with you — you can edit and run this or — view only at the top.
• Give them their own copy — a separate workflow in their account that they can change freely. Choose whether they share the same tables (one source of truth, kept in sync) or get their own copy of the tables, and whether to Include workflow outputs. Only view in apps limits them to using it inside Apps.

Later, Send my changes to all pushes your current version to everyone who has a copy; they see it as an update they can apply or keep their own version (Keep my version). If you received a copy, Sync shared workflow pulls the owner's latest version.

Sharing is disabled while the workflow is running.

Appify

Appify publishes a workflow to Apps so that people can run it without seeing the canvas. Choose the app type:

• Generic App — a form of the fields you marked editable in app. Fill it in, run, and every run appears as a timeline. Works with any workflow.
• Chatbot App — a conversation. The workflow must start with a User Input or Super Agent node; the agent keeps chat history.

You will be asked to choose which nodes export their output (that is what app users see as the result) and to write a short description of what the app expects and returns. Unappify takes it down again.

Tags

Attach Workflow Tags from the info toolbar to organise your list; Manage Tags creates, renames or deletes them. Tags also filter the Apps hub.

Trash

Delete moves a workflow to the trash for 30 days, from where it can be restored. Deleting shows exactly what else goes with it — generated files, dashboard charts, chat conversations opened on it, run history, schedules, queued webhook runs — and what stays (tables it merely reads belong to your data and are kept). People who received their own copy keep it; people it was shared as an app with lose access. Delete permanently removes it before the 30 days are up.

Do it with the API

Everything above has an endpoint: creating and saving specs, running and stopping, outputs and versions, schedules, sharing, tags, and the webhook trigger. See Workflows, Workflow assistant and Webhooks in the API reference, and Streaming for live node status over WebSocket.