Skip to main content
The copilot is an assistant built into the studio that can see your whole workflow. Ask it to build a pipeline from scratch, change a setting, debug an error, or explain what a graph does — in plain language, without touching a node by hand.

​
Two ways to reach it

  • The chat panel — open it from the studio to have a full conversation. It shows the copilot’s replies, the changes it proposes, and any run results.
  • The input bar — a compact “Ask the copilot” bar sits above the canvas controls. Type a request there and press Enter; the panel opens and your message is sent. The bar expands as you focus it so you have room to write.
On an empty canvas, the panel’s starting screen offers a few templates to build from. On a populated canvas, it offers prompts to explain, debug, or improve what’s already there.

​
Build a workflow

Describe what you want and the copilot assembles the nodes and connections for you.
1

Describe the outcome

For example: “Build a workflow that turns a brand brief into three social images.” Be as specific or as loose as you like — the copilot fills in sensible defaults.
2

Review the proposed build

The copilot shows the workflow it plans to create. Nothing lands on your canvas until you approve it.
3

Apply it

Approve the change and the nodes appear, wired and laid out. Reject it and your canvas is untouched.

​
Edit an existing workflow

The copilot makes surgical edits — it changes only what you asked for, rather than rebuilding the graph. “Change the prompt to a red sneaker on white” or “switch this node to a different model” each land as a single, targeted change with the rest of the workflow left alone. It reads your current graph directly, so it already knows each node’s settings and current values before it proposes a change.

​
Suggest and Auto

An Auto toggle in the input controls how proposed changes are applied.
  • Off (Suggest) — the default. Every change is shown as a card that you Apply or Dismiss, so you review before anything happens to your canvas.
  • On (Auto) — changes apply automatically as the copilot makes them, for a faster back-and-forth.
Even in Auto, deletions always ask for confirmation. The copilot never removes a node or connection without a card you approve.
Applied changes are undoable with the usual undo shortcut.

​
Run, see, and refine

When you want a finished result rather than just a graph, ask the copilot to run the workflow. It executes the run, looks at the output, and judges it against your request. If it’s off, it refines a setting and runs again — up to a few times before pausing to ask whether to keep going. See Running agents for how runs execute.

​
Give it context

Point the copilot at specific nodes with an @ mention in the input. A node picker opens; pick the node (image and video nodes show a thumbnail) to attach it as context so the copilot focuses its answer there. You can also attach images to your message for the copilot to look at.

​
Brands

If your workspace has brand kits, the copilot uses them when you ask for on-brand media.
  • With one brand, it applies that brand’s palette and voice and tells you it’s doing so.
  • With several brands (for example, an agency with multiple clients), it asks which brand to use before generating, then applies the one you pick.
It weaves the brand’s colors and tone into the prompts and settings it writes, and can add a brand-kit node wired into your generation nodes so the brand carries through at run time.

​
Chat history

Conversations are saved per agent. Refresh the page or come back later and the panel resumes your last conversation where you left off. To revisit an older conversation, open the history view from the chat panel’s header. It lists your past conversations for the agent with titles and timestamps — select one to continue it, use New chat to start fresh, or Back to return to where you were.

​
Memory

The copilot can remember durable preferences across chats — for example, “I always use a 16:9 aspect ratio.” When you state a lasting preference, it’s saved quietly and recalled in future conversations so you don’t have to repeat yourself.
Memory is optional and only active when your workspace has it configured. Without it, the copilot works normally but doesn’t carry preferences between chats.

​
Credits

Copilot chat calls a language model, so each turn draws from your workspace credits, the same pool your runs use. If your balance is empty, the copilot pauses and prompts you to top up before continuing. See Credits for how the credit pool works.

​
Explain and debug

Beyond building, the copilot reads and reasons about your graph:
  • Explain — “Walk me through this workflow” describes the path from inputs to outputs, node by node.
  • Debug — “What’s wrong with my workflow?” checks for validation errors and points at the node and reason.
  • Improve — “How can I make this better?” suggests concrete changes you can apply.