AI features (bring your own key)
FlowBot's runtime is deterministic by default — AI only enters your bot where you explicitly add it, and always with your own API key. Supported providers: Anthropic (Claude), OpenAI and OpenAI-compatible endpoints (Groq, OpenRouter, Mistral… via a custom base URL), and Google Gemini.
AI Builder — describe a bot, get the flowchart
Open ✨ AI Builder in the header, choose a provider, paste your key (it's stored only in your browser's localStorage, never on FlowBot's servers), and describe what you want: "A bot for my Italian restaurant in Pune — menu, table booking, opening hours and AI support." The assistant draws the whole flowchart on your canvas, inventing custom blocks where no built-in fits. Keep chatting to refine it; ↩ Undo restores the flow from before the last AI edit. Every generated flow is validated before it reaches your canvas.
AI Reply block — open-ended chat inside a flow
Drop an AI Reply block anywhere (e.g. behind a "Talk to our assistant" menu option) and configure:
- Provider + API key — stored with your bot's flow server-side, used only server-side to call the provider; never sent to visitors and stripped from share-page clones.
- Business context — the system prompt: who you are, tone, facts it may use.
{vars}work here too. - Model (optional) — defaults:
claude-haiku-4-5,gpt-4o-mini,gemini-2.5-flash.
While the block is active, every customer message goes to the LLM with the last few turns of context (replies are capped at 500 tokens). The customer types 0 — or "exit", "menu", "back" — to leave AI chat and continue the flow. If the provider errors, the bot sends your fallback message instead of going silent.
AI step in custom blocks
The Block Lab has the same capability as a step — AI chat until 0, then the remaining steps run. Use it to build features like "AI answers questions about the menu, then asks for the order".
Cost & privacy notes
- You pay your AI provider directly at their rates; FlowBot adds no markup and no quota.
- Deterministic parts of your flow (orders, prices, bookings) are untouched by AI — the LLM only sees the conversation inside its own block.