ai-chat-gated

Optional module — streaming AI chat where each message costs sats. Invoice, WebLN payment, then Vercel AI SDK streamText.

Metered AI without subscriptions. Each user message triggers a Lightning invoice; after WebLN payment and preimage verification, the server streams an LLM response.

Files added

File	Purpose
`lib/chat-payment.ts`	Invoice creation, payment verification
`lib/ai/providers.ts`	Vercel AI SDK provider factory
`components/ai/PaymentGate.tsx`	Pay-to-send wrapper
`components/ai/GatedChat.tsx`	Chat UI with payment gate
`components/ai/ChatMessage.tsx`	Message bubble component
`app/api/chat/route.ts`	POST — stream AI after payment check
`app/api/chat/invoice/route.ts`	POST — create per-message invoice
`app/demo/ai-chat/page.tsx`	Demo page

Dependencies

ai, @ai-sdk/react, react-markdown
@ai-sdk/anthropic, @ai-sdk/openai, @ai-sdk/groq, ollama-ai-provider

Environment variables

Key	Required	Description
`AI_PROVIDER`	yes	`anthropic` \| `openai` \| `groq` \| `ollama`
`AI_API_KEY`	yes*	Provider API key (*not for Ollama)
`AI_MODEL`	no	Model name override
`AI_BASE_URL`	no	Ollama/self-hosted endpoint
`AI_SATS_PER_MESSAGE`	no	Server-side price (default from env)
`NEXT_PUBLIC_SATS_PER_MESSAGE`	no	Client UI display price

Flow

1. User types message in GatedChat
2. PaymentGate fetches POST /api/chat/invoice
3. User pays invoice via WebLN
4. Client sends POST /api/chat with messages + payment proof
5. Server verifies payment, calls streamText(), returns SSE stream
6. GatedChat renders streaming response via @ai-sdk/react useChat

API routes

POST /api/chat/invoice

Returns a BOLT11 for one message at AI_SATS_PER_MESSAGE sats.

POST /api/chat

Accepts chat messages and payment verification headers. Returns streaming AI response on valid payment.

PaymentGate

Wraps message submission:

<PaymentGate onPaid={(proof) => submitMessage(proof)}>
  {(requestPayment, isPaying) => (
    <ChatInput onSubmit={requestPayment} disabled={isPaying} />
  )}
</PaymentGate>

Dev bypass

x-mock-preimage header in development skips WebLN for API testing.