# I launched two L402 APIs for AI agents — clean data for 10 sats, mempool forensics for 2,500

> Source: <https://dev.to/cramberries099bit/i-launched-two-l402-apis-for-ai-agents-clean-data-for-10-sats-mempool-forensics-for-2500-1c5d>
> Published: 2026-10-22 19:07:24+00:00

AI agents can write code, browse the web and run tools — but the moment they need to *pay* for an API, the flow collapses into human infrastructure: signups, API keys, credit cards, monthly plans.

L402 fixes that. The payment **is** the credential. So I built two production-light APIs that any agent (or human) can consume autonomously over Bitcoin Lightning, and put them live:

**🔗 [https://l402-ai-gateway.onrender.com](https://l402-ai-gateway.onrender.com)**

| Endpoint | Price | What it does | 
|---|---|---|
| `POST /v1/data/clean-scrape` | **10 sats** | Takes raw web content and returns clean, structured JSON an LLM can consume directly | 
| `POST /v1/forensic/mempool-arbitrage-audit` | **2,500 sats** | Analyzes a real Bitcoin transaction (fee rate, RBF signaling, inputs/outputs, confirmations) using live mempool.space data, and compares live BTC prices across Kraken, Coinbase and Bitstamp to flag arbitrage spreads | 

No accounts. No keys. You call, you get a `402`, you pay the Lightning invoice, you retry with the proof, you get the data.

The middleware is protocol-standard, not a custom hack:

`402 Payment Required` with a `WWW-Authenticate: L402 macaroon="…", invoice="lnbc…"` header (plus the same data as JSON).`Authorization: L402 <macaroon>:<preimage>` → the server verifies `sha256(preimage) == payment_hash` locally and serves the result.
It works out of the box with standard tooling like Lightning Labs' `lnget` — point it at an endpoint and the pay-and-retry cycle is automatic.

Discovery is built in too: a machine-readable manifest at `/.well-known/l402.json`, public pricing at `/pricing`, interactive OpenAPI docs at `/docs`, and the service announces itself on Nostr with NIP-89 handler events (kind 31990) so agents can find it without me doing marketing.

Under the hood: FastAPI, an in-memory TTL cache (repeat calls are cheap), and live upstream data — nothing in the forensic report is mocked.

This is a launch, not a victory lap:

The cheapest call costs 10 sats — less than a cent. If you're building agents, trading bots or L402 tooling, I'd genuinely love for you to:

Docs and manifest are public, the code structure is a standard FastAPI service, and the paywall answers with a real invoice from a real wallet. The machine economy needs services that are actually *buyable* by machines — this is my first one.

What should the third endpoint be?