AI Assistant

Every board has a built-in chat assistant in the side panel. It can answer questions about Banana Flow, explain what’s on your canvas, and — uniquely — directly create, update, and connect nodes for you.

What it can do

• Add nodes: “Add a prompt node and an output node”
• Wire connections: “Connect that prompt to the new output”
• Pick models: “Set the output node to GPT Image 2”
• Generate: “Run the generation with the current setup”
• Iterate: “Make the prompt more cinematic and re-run”
• Explain: “Why did this generation fail?”

The assistant uses an LLM (via Vercel AI Gateway) and has live access to your board state, so it can act on what’s actually there rather than guessing.

When it’s useful

• Onboarding — let it set up your first flow while you describe what you want
• Bulk edits — “rename all my prompt nodes to use ‘photorealistic’ instead of ‘realistic’”
• Exploration — “What models are good for portrait photography?”
• Debugging — “Why is this output node greyed out?”

When to drive manually

For fine-grained editing (dragging a node, picking exact aspect ratios, tuning the prompt word by word), it’s still faster to do it yourself. Treat the assistant as a co-pilot, not an autopilot.

Limits

• Each chat message costs a small amount of credits (much less than a generation)
• Long-running generations are still gated by your normal credit balance