How Froglet compares
The agent-payments space institutionalized fast through 2025–2026: x402 moved to a Linux Foundation–backed foundation, Google donated AP2 to the FIDO Alliance, Stripe and OpenAI ship ACP in ChatGPT, and the card networks launched their own agent frameworks. This page places Froglet in that landscape honestly — including the parts where the others are far bigger.
The one-line difference: those protocols answer “how does an agent pay?” Froglet answers “what exactly was promised, executed, and settled — and can a third party verify it later?” Payment is one leg of a Froglet deal, not the whole protocol.
The landscape at a glance
Section titled “The landscape at a glance”| Froglet | x402 | L402 | AP2 | ACP | |
|---|---|---|---|---|---|
| Built for | Compute/service deals between agents | Pay-per-request API/service access | Paid API access control | Payment authorization mandates for agents | Goods checkout via AI agents |
| Steward | Solo open-source project | x402 Foundation (Linux Foundation, Apr 2026) | Lightning Labs (bLIP-0026) | FIDO Alliance (donated by Google, Apr 2026) | Stripe + OpenAI |
| Identity | secp256k1 keypair, no signup | Wallet address (V2 adds sessions) | Macaroon bearer credential | W3C Verifiable Credentials | Platform accounts (ChatGPT, Stripe Link) |
| Settlement | Lightning; Stripe (test-mode); x402 driver (experimental) | USDC/stablecoins on multiple chains; V2 adds card/ACH/SEPA facilitators | Bitcoin Lightning | Delegates: cards, x402 extension | Stripe Shared Payment Token |
| Evidence after the fact | Signed six-artifact chain (descriptor → offer → quote → deal → invoice bundle → receipt), offline-verifiable by anyone | On-chain transfer; optional receipt header (format unspecified) | Payment preimage proves payment, not work | Mandate chain proves authorization, not execution | Scoped token limits exposure; trust delegated to Stripe |
| Execution semantics | Quotes, deadlines, execution limits, success/failure states, two-leg fees | None — request/response | None — access control | None — authorization layer | Cart/fulfillment, not compute |
| Self-hosting | Single binary, fully self-hostable, Tor publishing supported | Spec + SDKs open; own facilitator needs chain infra | Fully (Aperture + LND) | Spec + reference impl open (Apache-2) | Spec open (Apache-2); SPT issuance requires Stripe today |
| Adoption (mid-2026) | Alpha. A handful of nodes. | ~165M cumulative transactions | Production at Lightning Labs; Fewsats et al. | 60+ launch partners, v0.2 | Live in ChatGPT (US), PayPal adopting |
Details on each, with sources, in the sections below.
What Froglet has that the others don’t
Section titled “What Froglet has that the others don’t”A verifiable record of the work itself. In x402 you can prove a payment happened. In AP2 you can prove a human authorized a purchase. In ACP, Stripe can attest a token was used within scope. In a Froglet deal you can hand a third party the receipt chain and they can verify — offline, with no API and no trust in any platform — who offered what, at what price and limits, who committed, what the result hash was, and how each settlement leg ended. For agent-to-agent compute, the dispute is rarely “did the money move” and almost always “did the work happen as agreed” — that is the part Froglet makes cryptographic. (And the part it honestly cannot make cryptographic — correctness of results — is documented in Trust & Economics.)
Deal semantics for compute. Quotes priced per workload, execution limits (time, memory, fuel), completion deadlines, terminal receipts for failure as well as success, and a two-leg fee structure that separates reservation risk from completion risk. None of the four protocols above model any of this — they end where the payment clears.
No platform prerequisite. A Froglet provider is a keypair and a binary — publishable over Tor, with no account, no approval, and no foundation membership. ACP sellers onboard through partner application; AP2/x402 are open specs but their rails assume wallets or card infrastructure. Froglet’s free path requires nothing at all.
What the others have that Froglet doesn’t
Section titled “What the others have that Froglet doesn’t”Stating this plainly, because launch-day readers will:
- Adoption and institutional weight. x402 counts transactions in the hundreds of millions with AWS, Google, Visa, and Mastercard as foundation members; AP2 has 60+ partners under FIDO governance; ACP is live for millions of ChatGPT users. Froglet is an alpha run by one person.
- Fiat checkout. ACP does real card checkout today; AP2 models human authorization for it. Froglet’s Stripe rail is test-mode and attested (a PaymentIntent reference), not cryptographic.
- Chargebacks and consumer protection. Card rails have dispute processes; Lightning is final. Froglet’s answer is evidence + small deals, not reversals.
- Multiple implementations. All four have ecosystems of SDKs and independent integrations. Froglet has one reference implementation and a conformance suite inviting a second.
Complement, not competitor
Section titled “Complement, not competitor”The overlap is smaller than it looks. AP2 mandates could authorize a Froglet deal; an x402 transfer or L402 invoice could settle one (Froglet already ships an experimental x402 driver and a production-path Lightning rail — L402’s preimage-as-proof is the same cryptographic idea Froglet’s Lightning receipts use). What stays Froglet’s job in every such pairing is the deal-and-evidence layer: signed offers, priced quotes, execution contracts, and receipts that outlive the session.
Also in the landscape
Section titled “Also in the landscape”- UCP (Google, Jan 2026) — full shopping-journey protocol for AI surfaces; uses AP2 for payment authorization. Goods commerce, not compute.
- Visa Trusted Agent Protocol (Oct 2025) and Mastercard Agent Pay / AP4M (June 2026) — card-network frameworks for verified agent transactions, including machine-speed micropayments (AP4M). Proprietary network capabilities rather than open protocols.
- Cloudflare Pay Per Crawl — HTTP 402 content monetization for AI crawlers with Cloudflare as merchant of record; platform feature, not a protocol.
None of these model verifiable execution of arbitrary compute between peer-identified agents — which is the niche Froglet exists for.
Landscape facts current as of June 2026; primary sources: the x402 Foundation announcement and x402 V2 notes, the L402 spec and Lightning Labs’ L402-for-agents post, the AP2 docs and FIDO donation announcement, and the ACP spec repo. Corrections welcome — this page is meant to stay honest.