Overview

FlowFn includes AI-powered assistance features to help you build workflows, playgrounds, forms, visualizers, agents, and streams faster. The assistant — Fen, the FlowFn mascot you'll see around the dashboard — generates code, drafts payloads, and answers questions about your setup.

Starting a chat — the setup wizard

Open AI Assistance from the dashboard sidebar. The first time you start a chat the assistant walks you through a short setup so it knows what to build and which tools to consider:

Pick what you're building. Workflow, playground, form, visualizer, agent, or stream. The assistant tailors its instructions and response shape to that one kind.
Pick your platform integrations and AI providers. Workflows, playgrounds, agents, and streams show a two-column picker: search and tick up to 10 platform tools (Slack, Stripe, etc.) and up to 3 AI providers (OpenAI, Anthropic, Gemini, …). Forms and visualizers skip this step because they don't wire to tools directly.
Describe what you want. A short textarea with click-to-fill examples tailored to the artifact kind.
Start with AI. The chat opens with your brief already sent and the assistant starts working immediately.

Your setup choices are remembered for the session, so reloading the page returns you to the same chat. A Start over button at the top lets you discard the setup and re-enter the wizard.

Persistent conversations

Every workflow-chat session is saved as a conversation in the right-side panel. Click a row to reopen it and the chat reloads with the full message history (the last ten turns load first; scroll up to fetch more in batches of ten). Useful patterns:

Switch threads while the AI is working. If you start a new chat with Start over while a turn is still being processed, the original conversation keeps running in the background. Click it later and you'll see the spinner — or the finished answer — without losing context.
One in-flight turn per conversation. If you click Send twice in a row, or open the same conversation in two tabs and send from both, the second attempt is held back with a "previous turn still in flight" message until the first one finishes.
Delete a conversation. The trash icon on each row soft-deletes the thread, which then disappears from the panel immediately. Soft-deleted conversations are permanently removed 60 days later. We do not let you delete a conversation while a turn is mid-flight — wait for the response.

One-shot artifact generations (workflow templates from a single prompt, code generation, etc.) are listed separately under View all, not as conversations.

How the assistant uses your tool selection

Pre-selecting tools keeps the assistant focused and the input prompt small, which makes turns faster and cheaper. If the assistant needs a tool you didn't pre-select, it can look up the full catalogue mid-chat (you'll see a short status like "Looking up 'whatsapp'…") and add the tool to your selection automatically. Two ways the selection can grow without leaving the chat:

Auto-added by the assistant. Any tool the assistant ends up referencing in your workflow / playground is added to your selection on the same turn — the toolbar pills above the chat update to reflect the new count.
Added by you with the toolbar. Click + Integrations or + AI providers in the chat header to open a side drawer and widen the selection. Saving applies the change to the next turn.

Working with the assistant's output

When the assistant produces a workflow, playground, form, visualizer, agent, or stream, it shows a preview card at the bottom of the chat with two actions:

Create. Opens a confirmation dialog where you can rename the artifact and save it to your team.
Refine. Collapses the preview, dismisses any pending selection panel, and pre-fills the chat input with "I'd like to change: ". Type your change and send to iterate.

The preview card has a chevron to collapse / expand at any time — useful for very tall previews when you're reading the conversation. Each new generation auto-expands the card so you don't miss fresh output.

Code generation in workflow builder

Code tasks in the workflow builder include an AI code generation option. Describe what you want the code to do in plain language, and FlowFn generates JavaScript code for your task. You can refine the prompt and regenerate until the code meets your needs. Always review and test generated code before publishing your workflow.

Usage and credits

AI Assistance and builder code generation use the same user credits as workflow runs when your plan applies a charge for platform-managed AI. After a successful model response, credits are debited using your plan's rules (often including a token-based rate). If your balance is too low for that charge, the product may block the request until you add credits or connect your own API keys where supported.

AI Model tasks inside published workflows follow the same idea: steps that run on your registry connections use your provider account; steps that run on FlowFn's shared keys are included in workflow run credits as platform AI usage. See Billing and Plans and Using AI Models for how limits and pricing appear on your plan.

Privacy and retention

Each AI request is recorded as an audit row containing your prompt, the model response, your wizard selections (artifact kind and pre-selected tool codes), and token counts. Image attachments are stored as metadata only. Audit rows are purged automatically once they exceed your plan's retention window — windows step up with tier and again for annual cycles (Free / Student 2 months, Starter 3–6 months, Pro 6–12 months, Business 12–24 months).

Conversations. When you delete a conversation it is hidden from the right-side panel immediately and permanently removed after a 60-day grace window. The underlying audit rows are retained on the schedule above so billing reconciliation and compliance audits keep working. You can object to AI processing entirely from My Account → Privacy Rights — pick AI assistance as the processing type. See Security and Privacy.