# satgate - Lightning-paid AI inference

> This endpoint provides OpenAI-compatible inference behind HTTP 402 payments.
> Pay with Lightning, LNURLcash (LUD-25) bearer notes. No account required.

## Available Models
- gemma3:4b (30 sats / 1k tokens)
- qwen3:0.6b (10 sats / 1k tokens)

## Pricing
Requests are billed per token (prompt + completion) at the rates above. The worst case for max_tokens is held up front and the unused part refunded once the upstream reports usage.

## Payment Schemes
- L402 (credit-based: pay once, spend the balance over many requests)
- IETF Payment (per-request, draft-ryan-httpauth-payment-01)

When payment is needed, the server returns HTTP 402 with a challenge for each scheme it accepts. Clients can choose any of them.

## Usage
Send standard OpenAI-compatible requests to /v1/chat/completions, /v1/completions or /v1/embeddings.
The first request returns 402 with payment details. Pay, then retry with your credential:
- L402: Authorization: L402 <macaroon>:<preimage>
- IETF Payment: Authorization: Payment <base64url credential>
- LNURLcash: X-LNURLcash: <note URL>

## Buying credit
POST /create-invoice to get a Lightning invoice for credit up front.

## IETF Payment Authentication
Implements draft-ryan-httpauth-payment-01 with the lightning payment method.
See: https://github.com/forgesworn/payment-methods