Skip to content
Docs
flero.ai

AI builder

Flero's built-in assistant for building workflows from a description. The floating chat bubble in the bottom-right of the editor opens it.

📸 Screenshot needed: editor__ai-builder-chat.png, floating chat panel open with an example prompt typed in and a suggested workflow rendered in the chat.

What it can do

The AI builder is wired into the workflow editor and aware of:

  • Every available node type and its inputs / outputs.
  • The current workflow on your canvas (nodes, connections, configurations).
  • The connectors you have credentials for.

You can ask it to:

  • Generate a new workflow from a description.

    "When a new order is created in Shopify, send a Slack message to #orders with the customer name and order total."

  • Add or extend a step in the current workflow.

    "After the HTTP Request, add a step that uses an AI Agent to summarise the response in two sentences."

  • Explain part of the workflow to you.

    "What does the loop node here actually do?"

  • Suggest fixes for a failing run.

    "The Slack node is failing, what's the most likely cause?"

It cannot run code, change credentials, or take actions on your behalf, every suggestion lands as a proposal you accept before it modifies the canvas.

How to use it

  1. Click the chat bubble (bottom-right of the editor) to open the panel.
  2. Type a request in plain English.
  3. The assistant responds with either:
    • A textual explanation, or
    • A proposed change rendered as a diff (e.g. "Insert AI Agent between HTTP Request and Log; configure prompt to …").
  4. For proposed changes, you'll see two buttons: Apply and Reject.
    • Apply inserts the nodes and connections into your canvas. You can still inspect and edit before saving.
    • Reject discards the proposal but keeps the chat history for context.

Best practices

  • Be specific about inputs and outputs. "When a Shopify order arrives" is better than "after a sale". The assistant uses these hints to pick the right trigger.
  • Name credentials in your prompt. "Use my Shopify credential called Prod store", the assistant will wire it in directly.
  • Iterate. Apply, run, ask for adjustments. Treat each accepted suggestion as a checkpoint.
  • Cap the scope. Asking for a 30-node workflow in one shot rarely works. Ask for the trigger and the first action; build up from there.
  • Review every applied change before running for the first time. The assistant occasionally hallucinates fields that don't exist on a node; you'll see them as red-bordered nodes with validation errors.

What it costs

The AI builder calls an LLM under the hood. Token usage shows up in your Settings → Billing & usage under "AI assistance". A small workflow tweak is typically a few hundred tokens; large generations can be a few thousand.

If your workspace has set a Token cost guard (see Settings → Billing), the assistant respects it and will pause if you exceed your budget.

Tips & gotchas

  • The assistant doesn't always know about newly installed plugins. If you've just installed a marketplace plugin, its nodes may not appear in suggestions for a short window, refresh the page.
  • Chat history is per-workflow. Open a different workflow and the chat starts fresh. (Settings → Account → Privacy → Clear AI history wipes everything.)
  • You can switch models. The drop-down at the top of the chat panel lets you pick between Claude, GPT, Gemini, or whatever your workspace administrator has enabled.

Found something out of date? This page lives in the Flero docs content set.