AI Assistant

The AI assistant is the chat panel on the right of the workflow editor. You describe the automation in plain language and it composes the workflow on the canvas - nodes, connections, prompts and tool settings; later it adapts the workflow to your case, explains the workflow, reviews it for problems and debugs a failed run. One panel, one conversation: the assistant tells which of these you mean from the message you send. (The App Assistant is its counterpart for forms, pages and custom apps.)

Opening the assistant

Click the Sparkles icon (AI Assistant) in the editor toolbar to open the chat panel on the right side of the canvas. It uses the same slot as the Inspector and the Executions panel - only one is visible at a time, and switching keeps the conversation, including a reply that is still being written.

The panel also opens by itself in two situations:

• New blank workflow - the chat opens in build mode with the prompt "Let's build your workflow" and the input placeholder "Describe your workflow…". Describe your goal; the assistant first shows a plan card summarizing what it intends to create - click Build it to let it start adding nodes. You can also just close the panel and build manually.
• Workflow started from a template - the chat opens with the template's text already in the input. Edit it for your case and send it; the assistant treats it like any build request. Nothing is sent automatically. See Templates.

Building a workflow from scratch

1. Create a new workflow and choose Start from scratch. The editor opens with the assistant already active: "Let's build your workflow".
2. Describe what you want to automate. The assistant grounds itself in your workspace (the AI models, MCP servers, knowledge bases, apps and global variables you have), so it only proposes things that can actually run. If something your request needs is not configured yet - say, Slack - it says so and offers an inline setup form instead of building a broken flow.
3. If it needs more information, it asks one focused question - answer and continue.
4. It shows a plan card summarizing the workflow it intends to create. Review it and click Build it.
5. Nodes and connections arrive as action cards - accept them one by one or Apply all. In Manual apply mode nothing changes without your confirmation; in Auto-apply the cards apply as they arrive.
6. Before calling the workflow ready, the assistant checks its requirements. Anything missing - an AI provider key, an MCP server, a global variable - appears as a setup card you fill in without leaving the chat.

The assistant builds the workflow but does not run it: review the nodes, switch the workflow to Active and press Run. A very branched flow may need a second pass - build the core first, then iterate node by node in the same chat. To build the form or page that starts a workflow, use the App Assistant - it can create the workflow for you, too.

Tips for good prompts

• Be specific about the trigger: "Create a webhook that receives order data…", "Run every day at 9am…", "Manual trigger that accepts a list of URLs…".
• Describe the data flow: "Fetch the stock prices, then analyze trends with AI, then send a summary to Slack"; "Receive a support ticket, search the knowledge base, generate a response".
• Mention the output format when it matters: "Return JSON with fields status, summary and action_items"; "Send a formatted report to the #analytics Slack channel".

Prompts that work well:

• "Create a workflow that monitors stock prices for AAPL and TSLA every hour. If any stock moves more than 5% from the previous check, send a Telegram notification with the details."
• "Build a webhook handler that receives customer feedback, analyzes sentiment using AI, and stores the result in a Google Sheet. Use the knowledge base to find relevant product documentation for context."
• "Make a daily scheduled workflow that fetches open YouTrack issues, groups them by priority, generates a summary report with AI, and posts it to the #dev-updates Slack channel."

What else the assistant can do

You can switch modes mid-conversation - the assistant follows the latest user intent rather than locking into a behavior.

Workspace awareness

Every reply is grounded in the resources actually configured for the current workspace:

• AI Providers - only models from connected providers are suggested
• MCP Servers - only tools from servers assigned to this workspace; full param schemas are loaded
• Knowledge Bases - surfaced for Agent context or rag Action nodes
• Apps - the workspace's forms and pages, as App trigger candidates with their declared fields
• Global Variables - referenced via {{global.<key>}}
• Previous run output - so {{steps.<id>.output.<field>}} references point at fields that actually exist

If a required resource is missing, the assistant emits a setup card with an inline form instead of telling you to leave the chat: an AI provider key (pick the provider, paste the key), an MCP server (a form generated from the server's config schema), or a global variable (value + Save) - everything is saved directly from the chat.

Where values go

When the assistant needs a value - a project key, an API base URL, a date range - it picks where the value belongs by how often it changes; see Variables > Where a value belongs. When that is unclear ("is this workflow always for one project, or many?"), it asks you one question before committing.

Action cards (Accept / Apply all)

The assistant doesn't write changes silently. Every mutation it proposes - adding a node, updating a config, creating a global variable, requesting a missing setup - renders as a compact action card in the chat:

• The card shows the operation, the target node, and a diff/summary of what will change; expand a card to see the details and its Accept / Reject buttons.
• A card for a tool action shows the tool's parameters as typed fields with validation, and pre-fills references to the outputs of earlier steps.
• When the assistant emits multiple cards in one turn (e.g. "add three nodes and wire them up"), the chat shows an Apply all button at the top of the group with a "N pending changes" counter - one click applies everything in order, with structural changes batched and re-rendered as a single canvas update.
• Sensitive changes (e.g. touching credentials or destructive edits) carry an amber Review badge and are always expanded, so you see the diff before Apply all can include them - Reject the card if you don't want it.

By default (Manual apply mode) nothing is applied until you accept it, and you can ask the assistant to revise a proposal before accepting. If you'd rather not confirm each card, switch the header's Apply mode to Auto-apply: "the assistant's changes apply as they arrive; it only stops when it needs your input. Cmd+Z undoes; nothing runs until you hit Run." Questions and setup cards still wait for you, and so does a sensitive change written after the assistant read data from an external system - it is marked "Not applied automatically" for you to review.

If some cards in a batch cannot be applied (for example an edge to a node that was not added), the rest are applied and each failed card says why, with Retry.

Checks before "ready to run"

Before the assistant tells you a workflow is ready to run, it checks that everything it needs is actually in place. The check happens when it proposes the change, not after you apply it, so you don't accept a card and discover a missing prerequisite later.

It surfaces:

• MCP servers the workflow needs but that aren't configured or assigned to this workspace
• Tools that require credentials your organization hasn't set yet
• Global variables referenced in node configs but not yet in the workspace's Global Context
• An App trigger that names an app of another workspace, or one that no longer exists

Each item comes back as its own setup card with a working save form, so you can resolve everything without leaving the chat.

Quick starts

An empty chat on an existing workflow offers quick starts: Explain workflow, Analyze workflow and Debug last failure (on a blank workflow: What can you do?). They send the same requests as the matching slash commands below.

Checking a server while building

When you have both Create & Edit workflows and Execute workflows in the workspace (Owners and Admins always do), the assistant may read from a connected server while it builds - for example to use the real names of your report tables, columns or objects instead of guessing them. Each such check appears in the chat as one line, "Checked {tool} on {server}". It changes nothing in the external system, and the answers are not stored.

Attachments, Stop and Edit & resend

• Images - attach a screenshot (paperclip, paste or drop; PNG, JPEG or WebP, up to 3 per message), for example of an error or of the result you want. An image is sent with its message and the next two.
• Stop ends the current reply; changes already proposed stay as cards. Edit & resend puts your last message back into the input.
• Leaving the page does not cancel a reply: the assistant finishes it, and the editor shows a mark on the Assistant button and a notice when the answer is ready. A reply that arrived while you were away is not auto-applied - review its cards.

Slash commands

Type / in the input to open a command dropdown ("Type / for commands" is shown as a hint below the input); use Tab to autocomplete.

You can also trigger these implicitly - "review this for problems" works the same as /analyze.

Node context chip

Click any node on the canvas while the chat is open and it attaches as a compact chip above the input - the assistant then sees exactly that node as context for your next message. The chip has three one-click actions - Explain, Analyze, and Debug - scoped to that node, and an × button to detach it. The chip always reflects what the assistant will actually see: chip attached means the node is in context, chip dismissed means it isn't.

Usage limits

Assistant usage is metered per plan: each organization has a daily token budget (resets at midnight UTC) and each user a messages-per-hour limit. When either is exhausted the chat replies with a clear message about when it resets and what to do. Limits depend on your plan.

Members with read-only access to a workflow can still chat - the assistant answers questions and explains, but won't propose changes to the workflow.

Tips

• Reference nodes by name, not internal id: "update the Fetch Data node" beats "update node-3".
• Talk in any language - the assistant answers in the language you wrote the message in.
• One ask per turn beats five. The assistant intentionally proposes one or two changes per reply when adapting a workflow to your case; let it apply, see the canvas update, then continue.
• For build mode, start with the goal. "When a Slack message contains 'urgent', create a Jira ticket and assign it to me" gives the assistant enough to compose the graph end-to-end. "Improve my workflow" does not.

Related

• App Assistant - the assistant for forms, pages and custom apps
• Templates - ready-made descriptions the assistant builds workflows from
• Variables & Templates - where a value belongs and how to reference it