AI builder

Describe a workflow in plain language and Flero builds it for you. The AI builder is a chat that opens from the Flero bubble in the bottom-right corner of every page. It can also edit, explain and debug your workflows.

Opening the chat

• Click the round Flero bubble in the bottom-right corner. The AI builder opens as a panel docked on the right edge of the page.
• Drag the panel's left edge to make it wider or narrower.
• Click Minimize in the panel header to fold it back into the bubble.
• On an empty workflow, the canvas offers Ask flero and a few example jobs (Morning metrics, Lead → CRM, Competitor watch). Clicking an example opens the panel and sends it straight away.

The panel header also has New chat and History.

Connecting an AI model

The builder runs on an AI model from your own provider account. The first time you open it, a Connect your AI form asks for:

1. Step 1: Choose your AI provider: Anthropic, OpenAI, Google, or Ollama for local models where your deployment offers them.
2. Step 2: Add your AI provider credential: your API key. It is encrypted and stored as a credential in your workspace. Ollama needs no key.

After that, the provider and model dropdowns under the message box let you switch at any time. Only credentials from your current workspace are listed. If you pick a model that may not handle tool calls well, Flero warns you first.

What it can do

The builder asks you to confirm before it deletes a workflow or removes a node.

How a build works

1. Type your request in the message box and press Enter.
2. A status line shows what the builder is doing as it works, for example writing the workflow, checking it, or reading your run history.
3. The builder writes the whole workflow in one go and checks it before anything touches your canvas: node settings, connector operations, credentials and expression references. If a check fails, it fixes the problem and tries again.
4. Flero places the workflow on the canvas one step at a time, so you can watch it being built: each node, then the connections into it. A new workflow opens in the editor first.
5. The reply summarises what was built, with a Flow card that shows the finished graph.

Edits work the same way. The builder reads the current workflow and changes only what differs. Nodes you keep stay where they are, with their settings.

Click Stop (the send button while a turn is running) to halt a turn.

Agent and Plan modes

The mode dropdown under the message box has two settings:

• Agent (default): the builder makes its changes straight away.
• Plan: the builder proposes its changes first. They appear as a Proposed plan card with numbered steps. Click Apply all to make the changes, or Discard to drop them.

Setup steps after a build

Some things the builder can't do for you, such as choosing a credential or entering a secret. Under the latest build reply, a short list shows what is left: One thing to set up before this can run (or N things…).

The list shrinks as you finish each item and disappears when nothing is left.

Placeholder variables

When a workflow needs a secret that no credential can hold, the builder reads it from a workspace variable, {{ $env.NAME }}. If the variable doesn't exist, the builder creates it with the placeholder value SET_ME and lists its name in the reply. Set the real value before you run the workflow. A node that reads an unset workspace variable fails.

Attachments

Click the paper-clip button (Attach an image or document) or paste a screenshot into the message box.

• Supported: images (PNG, JPG, GIF, WebP), PDF, and plain text, CSV, Markdown and JSON.
• Limits: up to 8 files per message, 8 MB per file, 20 MB in total.

Chat history

Click History in the panel header to see your earlier chats. Each one has Rename and Delete. Click New chat to start fresh.

What it costs

Every message is one or more calls to the model you picked, paid with your own provider account. A build is usually one or two calls.

Workspace admins can set a monthly AI Builder (chat) budget in Settings → Workspace → AI Budgets. It is separate from the budget for AI Agent nodes (workflow runs), so one never blocks the other. The card also shows this month's spend so far.

When the builder budget is used up, the chat shows a message like Workspace monthly AI Builder budget exceeded: $X / $Y. and stops until the month rolls over or an admin raises the budget. Your workflows keep running.

See Billing & usage.

Best practices

• Name the trigger and the systems. "When a Shopify order arrives" beats "after a sale".
• Name credentials when you have several. "Use the Slack credential called Prod workspace."
• One workflow per message for big jobs. If a request is too large, the builder asks you to split it.
• Use Plan mode when you want to review changes to an important workflow before they land.
• Review before the first run. Work through the setup steps, then try a Test Run in sandbox mode if the workflow talks to real services.

Tips & gotchas

• Long turns can take minutes. A large build or a slow local model can take a while. The chat gives up on a single step after 12 minutes and tells you.
• "The AI request was rejected…" usually means the API key is wrong or expired, or the model name is not valid for that key.
• "Your AI provider authorization failed…" means you should reconnect the provider's credential.
• Local models are slower and less reliable at building workflows than hosted ones.

Related

• Editor overview: the canvas the builder draws on
• AI & LLM nodes: AI nodes you can ask the builder to add
• Managing credentials: the API key the builder uses
• Billing & usage: AI spend

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