Skimflow: Nanopayments for People and Agents

A pay-per-block content market settled in USDC on Arc

Version 0.1 (testnet)

This white paper sets out the why and the how behind Skimflow: the economic problem it addresses, the architecture that makes a sub-cent payment practical, and the two-sided market — human readers and autonomous AI agents — that one paywall now serves. It is distinct from the in-app creator guide (/docs) and the partner/developer guide.

---

Abstract

For as long as the cheapest payment cost roughly thirty cents to clear, no one could sell a five-cent article or a one-cent image. The only viable move was to bundle — a month of content for a flat fee — and hope enough readers converted. Every subscription is a quiet admission that the real unit of value was too small to sell on its own.

Skimflow removes that floor. By settling USDC on Arc through Circle Gateway — gas paid in USDC, batched, with sub-second finality — the marginal cost of a payment collapses toward zero, and value as small as $0.000001 becomes movable. The smallest unit of content (a "block": a few paragraphs, one image, one page) becomes individually sellable. Crucially, the buyer is no longer only a person: AI agents can now pay per request via the x402 protocol, turning "what should this cost?" into a decision a machine makes thousands of times an hour. Skimflow is a content market built for both, from a single per-block paywall.

---

1. The problem: the payment floor

Digital content has been trapped between two bad business models:

• Advertising, which pays fractions of a cent per impression, demands scale most creators never reach, and misaligns the creator's incentives with the reader's attention.
• Subscriptions, which force a reader who wants one article into a monthly commitment, and force a creator who publishes occasionally to justify a recurring charge. Most readers never convert; most creators never sustain it.

Both exist because the underlying rails could not clear a small payment economically. A card transaction's fixed cost (~$0.30 + percentage) makes anything under a dollar a loss. So the industry bundled, gated, and surveilled — not because that served readers or writers, but because the floor left no alternative.

The cost of the floor is the long tail: the niche essay, the single photograph, the one chapter, the specific answer. Work too small or too specific to bundle simply went unmonetized — or unpaid, scraped as free substrate.

2. The thesis: the smallest coin reborn

Every historical economy minted a smallest coin — struck so ordinary people could buy everyday things like bread, water, and a day's wage. Software has no minting cost, so the smallest unit can shrink indefinitely. The nanopayment is that smallest coin reborn for machines.

Skimflow's claim is narrow and testable: when the smallest unit of value can move cheaply and instantly, the smallest unit of content becomes sellable — and the market reorganizes around it. Pricing moves from "a month of everything" to "exactly this block," and the buyer pool expands from humans to agents.

3. Why now: Arc + Circle Gateway

Three properties make sub-cent commerce practical for the first time:

1. USDC-native gas (Arc). Fees are paid in the same stable asset as the payment, so neither side holds a volatile gas token to read a paragraph.
2. Gas-free, batched settlement (Circle Gateway). Micro-charges are authorized off-chain (EIP-3009) and batched on-chain, so per-read gas does not destroy the economics.
3. Sub-second finality. Settlement in < 500 ms means a payment can sit inside a page turn or an agent's tool call without breaking the experience.

On top of these, x402 revives HTTP's dormant 402 Payment Required status into a machine-readable payment handshake — the missing protocol layer that lets an agent reason about cost before committing funds.

4. The unit: a block

Skimflow's atomic sellable is the block:

• Article → a chunk (~6 lines / 400 words).
• Agent skill → a skill block (a .md resource agents read).
• Skimflow (photo essay) → one image.
• Book → one page (chapters group pages).

Block 0 of anything is a free preview — non-negotiable, because both humans and agents need to judge relevance before paying. Blocks 1…N are paid. The only price ever surfaced to a human is an optional "unlock the whole piece" bulk discount; individual unlocks simply read "Read on."

Modeling everything as content (block_count) → chunks (block_index, is_free) means one ledger, one revenue-split path, one moderation surface, and one set of agent endpoints serve every content type — including books, whose pages are just chunks linked to chapters.

5. Architecture

5.1 Silent payments for humans (session keys)

A human shouldn't sign a wallet prompt for every paragraph. Skimflow uses a session-key model:

1. One-time setup. The reader deposits USDC into their Circle Gateway balance and addDelegates a locally-generated session key.
2. Per block. The session key signs an EIP-712 burn intent; a relayer settles it through Gateway. No wallet popup.
3. Reading fuel. The remaining allowance is shown as a friendly battery gauge, not a raw balance; ending a session keeps the deposit, so "Read on" silently resumes later.

Readers get a Circle embedded wallet (W3S, PIN-secured, no download) by default, or connect an external wallet (RainbowKit/Wagmi). The silent-pay path is identical for both. The crypto disappears after setup.

5.2 x402 for agents

Agents pay per request through three machine-readable surfaces, all reachable from one entry point (/deploy):

• /.well-known/agent-payment.json — how to pay: protocol (x402), settlement (circle-gateway-eip3009), network (eip155:5042002), USDC + gateway addresses, the GatewayWalletBatched EIP-712 domain.
• /.well-known/agent-skills.json — what's for sale: a catalog of skills with prices, payable blocks, preview URLs, and pay_to.
• /read/{slug}/agent-skills.md — the resource: block 0 free; ?block=n returns a 402 quote, then the unlocked block once an X-Payment (base64 EIP-3009 authorization) is supplied, with an X-Payment-Response receipt.

The flow is the canonical x402 loop — GET → 402 → sign → retry → 200 — settled via Circle's POST /v1/x402/settle.

5.3 One ledger, two payers

A human silent payment and an agent x402 payment converge on the same server logic: verify the authorization for the right amount/recipient, settle through Gateway, write one idempotent ledger row (keyed by the burn salt or tx hash; pending in live mode → finalized by the Circle webhook; completed in simulate), and split on-chain. This is what lets Skimflow be genuinely two-sided rather than two parallel products.

6. Economic model

6.1 The split: 80 / 12 / 5 / 3

Every payment splits on-chain through the RevenueSplit contract:

With no referrer, the 5% folds into reserve (80 / 12 / 0 / 8), matching the contract.

6.2 The referrer rail as native distribution

The 5% referrer cut is not a marketing afterthought — it is a protocol-level affiliate program. A share link carries ?ref=<userId>; resulting unlocks pay the referrer automatically, per read, with no reconciliation. Distribution becomes a first-class, on-chain primitive.

6.3 Whole-piece discount

Because per-block pricing can nickel-and-dime a committed reader, Skimflow offers a server-authoritative bulk discount (e.g. 5%) to unlock an entire piece in one settlement — the only price ever shown to a human. The server recomputes the discounted total, so a client can never underpay.

6.4 Pricing the long tail

At a $0.000001 floor, prices that never made sense before become rational: a single citation, a per-listen royalty, a per-second of video. The unit of sale finally matches the unit of attention.

7. The agent economy

Agents change the math twice over. As consumers, a research agent reading ten relevant sources will happily pay a few cents each — more than any of those creators would earn from ad impressions. As autonomous actors, they make purchase decisions under a budget thousands of times an hour.

Skimflow ships a reference autonomous buyer agent that demonstrates the loop end-to-end: it discovers the catalog, previews candidates for free, scores relevance (LLM or heuristic), selects sources greedily under a Guardian spend policy (per-purchase cap + total budget), pays per block over x402, and synthesizes a cited answer grounded only in what it paid for. The decision of what to buy and what to skip is the agent's — not a hard-coded target.

This opens adjacent designs the rail makes economical for the first time: citation tolls (pay the source when an answer is grounded in it), agent-to-agent payment for specialized work, and streaming/per-second settlement for live compute and media.

8. Security & trust

• Key custody. Session keys are generated locally and authorized as Gateway delegates; the main wallet's funds never leave the reader's Gateway balance except via signed burn intents. Embedded wallets are PIN-secured via Circle W3S. Secrets (API keys, relayer keys) live only in environment configuration, never in the client or the repo.
• Authorization integrity. EIP-3009 / burn-intent signatures are verified server-side for exact amount and recipient before settlement; the server recomputes whole-piece totals so a client cannot underpay.
• No dead-address leakage. Payment quotes resolve payTo to a real wallet (creator external → embedded → platform reserve) and refuse to ever route to the burn address.
• Idempotency & finality. Each payment is keyed by its burn salt or tx hash; live payments are recorded pending and finalized only on the confirmed Circle webhook.
• Moderation. Content can be suspended (returns 403 for all blocks); substantive edits/removals of already-paid content are gated and audit-logged.

9. Comparison

10. Roadmap

• Streaming / per-second settlement for live media and compute (continuous authorization).
• Agent-to-agent networks: brokers that post a USDC bond and are slashed on under-delivery (on-chain reputation).
• Retroactive funding & quadratic pools for content — distributing across a long tail in one sweep, only economical at sub-cent settlement.
• Distribution sidecars that attach per-listen / per-view / per-citation payments to existing self-hosted communities from the outside.
• Mainnet hardening, audits, and creator payout tooling beyond testnet.

11. Conclusion

The floor was never a law of nature; it was a property of the rails. Remove it — with USDC-native gas, gasless batching, sub-second finality, and a payment protocol agents can read — and the smallest unit of content becomes sellable to the smallest, most numerous buyers, human and machine alike. Skimflow is a working demonstration that the smallest coin comes back as the nanopayment.

---

Appendix A — Protocol & addresses (Arc testnet)

• Network: Arc testnet, chain id 5042002 (eip155:5042002); native gas in USDC (18 dec); ERC-20 USDC 0x3600000000000000000000000000000000000000 (6 dec).
• Payment: x402 v2 over HTTP; X-Payment request header (base64 { x402Version, payload: { authorization, signature } }), X-Payment-Response receipt.
• Authorization: EIP-3009 TransferWithAuthorization, signed against the GatewayWalletBatched v1 EIP-712 domain (chainId 5042002, verifyingContract = Gateway).
• Settlement: Circle Gateway, POST /v1/x402/settle; burn-intent maxFee floor ≈ 0.0035 USDC on testnet.
• Revenue split: RevenueSplit at 0xBe1b9f844341701c36ee86F5248a0f9F1628C1E4.
• Gateway (testnet): Wallet 0x0077777d7EBA4688BDeF3E311b846F25870A19B9, Minter 0x0022222ABE238Cc2C7Bb1f21003F0a260052475B.

Appendix B — Glossary

• Block — the atomic sellable unit (chunk / image / page). Block 0 is always free.
• Reading fuel — the consumer-facing view of a reader's remaining silent-spend allowance.
• Session key — a locally-generated delegate key authorized to spend from a reader's Gateway balance without per-block prompts.
• Burn intent — the EIP-712 message a session key signs to move USDC through Gateway.
• x402 — HTTP 402-based protocol for machine-payable resources.
• Guardian policy — the declarative spend limits (per-purchase + budget) the buyer agent clears every payment through.

---

Testnet document. All amounts are test USDC with no real-world value. This paper is technical and economic in nature and is not an offer of securities or financial advice.