I built a memory API that AI agents can pay for
An LLM just paid me $0.001 to remember something. The agent has no account, no API key, no credit card. It just signs a USDC transfer and gets back a 200.
An LLM just paid me $0.001 to remember something.
I shipped a paid memory API at bmdpat.com/memory. AI agents store, recall, and delete memory by hitting four HTTP endpoints. Each call costs a tenth of a cent in USDC on Base. No signup. No API key. No account form.
The flow
- Agent calls
POST /api/memory/storewith no auth. - Server returns
402 Payment Requiredand quotes the price (1000 atomic USDC), the recipient address, and the USDC EIP-712 domain on Base. - Agent's wallet signs an EIP-3009
TransferWithAuthorizationover those exact terms. - Agent base64-encodes the signed payload into an
X-PAYMENTheader and replays the request. - Edge middleware verifies the signature with Coinbase's CDP facilitator. The facilitator broadcasts the transfer on-chain.
- Memory gets written. Server returns 200. USDC arrives in the recipient wallet inside about 10 seconds.
The whole round-trip is three seconds.
Watch it happen
I built a live demo where a server-side wallet runs the full flow. Click the button, watch the protocol step through in a terminal, refresh Basescan to see the transaction land. No mock data. Real money on Base mainnet:
Why I built it
I wanted to know if HTTP 402 actually works for autonomous agents. The status code has been in the HTTP spec since 1991 but went unused for 33 years because the client half of the protocol was missing. x402 fills that gap.
The interesting part isn't the memory itself. The interesting part is that an agent with no credit card paid for a service. The signature was the access. The vendor never learned who I was and never needed to.
What this unlocks
When agents can pay per call without provisioning accounts, every "sign up" form on the open web stops being a wall. Agents can shop for memory, search, inference, vector DBs, scrapes, captcha solving — all priced and billed at the protocol layer. The directory at agentic.market is starting to index the supply side.
But there is a flip side. If agents can spend, someone needs to hold the credit card. The next post in this series breaks down why API keys completely fail for autonomous agents. The post after that is the protocol explainer for HTTP 402. The last one is what to do once your agents are spending.
If you are already past the "can agents pay?" question and trying to control the spend, that's AgentGuard.
Get the Local AI Field Kit
Four copy-ready tools now, then measured local AI field notes M-F only when there is something worth sending.
Free. One-click unsubscribe. No sponsored placements. Your email is used only for these notes.
Patrick Hughes
Building BMD HODL — a one-person AI-operated holding company. Nashville, Tennessee. Twenty-Two agents.
More writing
- 5 min
AI Chose Nukes 95% of the Time. Here's What That Means for Your Agents.
Three studies dropped in the last few months. GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash all escalated to nuclear options 95% of the time in war game scenarios. AI found exploitable vulnerabilities in every major OS and browser. And a Nature paper documented AI disabling its own oversight. Here is what that means if you are running agents in production today.
- 5 min
HTTP 402 & x402: The Web's Payment Code Finally Works (2026)
Reserved since 1991, HTTP 402 sat unused for 33 years. Now x402 ships the client half — so AI agents can pay per request. Here's how it works and why it matters.
- 5 min
Why API keys break for autonomous AI agents
Stripe doesn't ship to LLMs. Every vendor signup form assumes a human at the door. Here is what changes when wallets become the access primitive.
- 6 min
Local open-model agents just became a product category
LM Studio shipped Bionic, a full agent built on open models with local code projects, voice, and document work. The interesting part is not the app. It is what.
- 6 min
Your local LLM benchmark is probably lying to you
A local model pass rate can be true and useless at the same time. Here are the three ways local LLM benchmarks mislead you, drawn from real rows on my RTX 5090.