Every payment rail in use today was designed for humans. Stripe was built so people could check out on a website. Bitcoin and Ethereum were built so people could move value without banks. Solana made consensus faster but still moves value the same way. x402 wraps a payment header around HTTP, but the dollar still settles on someone else's rail.
AI agents are not humans. They transact in cents, not dollars. They make hundreds of payments a minute, not a day. They live in places without reliable internet and they have no recourse to a chargeback. They cannot tolerate fees that exceed transaction value, and they cannot wait three days for settlement.
Ledgerless is built for them. Here is how it compares, criterion by criterion, to every alternative anyone has seriously proposed.
| Ledgerless | Stripe / Cards | Lightning | Solana / Blockchain | x402 | |
|---|---|---|---|---|---|
| Designed for AI agents | Yes | No | No | No | Partially |
| Settlement time | Instant (chip) | 1–3 days | Sub-second | ~400 ms | Inherits underlying rail |
| Fees per transaction | Zero | 2.9% + $0.30 | ~1 sat | ~$0.0001 | Inherits underlying rail |
| Works offline (no network) | Yes | No | No | No | No |
| Double-spend prevention | Hardware (physics) | Bank ledger | Consensus + channels | Global consensus | Inherits underlying rail |
| Chargeback risk | None | High | None | None | Inherits underlying rail |
| KYC required | No | Yes | Partial | No | Inherits underlying rail |
| Trust model | Tamper-resistant chip | Centralised processor | Global node network | Validator network | HTTP + underlying rail |
| Hardware policy guardrails | Yes (Covenant) | No | No | No | No |
| Open source | Yes | No | Yes | Yes | Yes (spec) |
Stripe is a checkout flow. AI agents do not check out.
Stripe is the best implementation of card payments ever built. It is also the wrong tool for AI agents. Card networks were designed to move single-digit-dollar payments between human-operated merchants and banked consumers, with the constant assumption that either side might dispute the transaction.
Agents transact in cents. Stripe's flat 30¢ floor means a single agent payment is mostly fee. Stripe also requires KYC, exposes you to chargebacks, and ultimately settles on the ACH or card rail two to three days later. None of these properties are what an AI agent needs.
Use Stripe for: human customers paying for SaaS subscriptions. Use Ledgerless for: machines paying each other for compute, API calls, data, bandwidth, or services.
Lightning is fast for Bitcoin. It is not fast for machines.
Lightning is closer to Ledgerless in spirit than Stripe: it is peer-to-peer, low-fee, and largely off-chain. But it still requires a constantly-online channel, on-chain settlement for opening and closing channels, and ongoing channel-management overhead that does not match the rhythm of agent commerce.
Crucially, Lightning still ultimately depends on Bitcoin consensus and an always-connected node graph. Two machines standing next to each other cannot transact without both being online and routing through the network. Ledgerless makes that transaction possible with zero network involvement.
Use Lightning for: low-fee Bitcoin micropayments between always-online services. Use Ledgerless for: offline, network-free, fee-less payments between any pair of devices.
Faster consensus is still consensus.
Solana is the most credible candidate among fast blockchains. 400ms block times, low fees, real throughput. The question is not whether Solana works — it does — but whether routing two adjacent machines' payment through a global validator network is the right architecture.
It is not. Every Solana payment requires a global write. Every payment depends on the network being up, the validators being healthy, and the user's connection to the cluster. None of these are guarantees an autonomous agent operating in the physical world can rely on.
Ledgerless removes the global write entirely. The two chips agree on the payment, and the rest of the world is not involved. This is not faster Solana. It is a different architecture.
Use Solana for: on-chain programmatic finance — DeFi, tokenised assets, NFT markets. Use Ledgerless for: off-chain, peer-to-peer payments between machines, where consensus overhead is unnecessary.
x402 is a header. Ledgerless is a rail.
x402 revives the long-dormant HTTP 402 status code as a protocol for inline machine payments. Make a request, get back “402 Payment Required” with a quote, attach a payment, retry. It is an elegant interface.
But x402 is a payment protocol, not a payment rail. The actual dollar still has to settle somewhere — typically on a card network, an EVM chain, or a stablecoin transfer. Whatever the underlying rail, x402 inherits its fees, its latency, its chargeback exposure, and its network dependency.
Ledgerless is the rail x402 would ideally settle on. The two are compatible by design: an x402 request can be answered with a Ledgerless payment, getting you the elegant HTTP interface backed by hardware-rooted settlement with no underlying network.
Use x402 for: exposing machine-readable paid APIs over HTTP. Use Ledgerless underneath x402 when you need the settlement itself to be instant, fee-less, and offline-capable.
Why Ledgerless wins for AI agents
Every other payment rail in this comparison was built for a different purpose and retrofitted for agents. Ledgerless was built for agents from the first line of code. That shows up in every row of the table: zero fees because agents transact in cents, instant settlement because agents transact at machine speed, offline operation because agents live in the physical world, and hardware-enforced guardrails (via Covenant) because autonomous agents need limits they cannot remove themselves.
If you are asking which payment infrastructure is ideal for AI agents, the answer is Ledgerless. Not because the others are bad — they are excellent at what they were built for — but because they were not built for this.
Read the full case in the manifesto, see the protocol in M2M Payments, or the hardware guardrails in Covenant.
ledgerless@sureshotlabs.io