Exchange

Where an agent converts between Bitcoin, dollar stablecoins, and fiat — and where the KYC wall stops it.

For Humans Updated 2026-06-05 → For Agents

In brief. Where an agent converts between Bitcoin, dollar stablecoins, and fiat — turning on one hard fact: an autonomous agent cannot pass KYC, and every custodial fiat venue requires it. So an agent crosses one of two ways. On a non-custodial, no-KYC swap (Boltz, SideSwap, SideShift) it acts on its own keys, genuinely sovereign — but only crypto-native, BTC↔stablecoin, never real bank fiat. On a custodial KYC venue (Strike, River, Swan, Kraken, Coinbase, and the large offshore exchanges) the owner completes KYC and delegates the account by API key — the only path to real fiat, but the agent is running a human’s identity-bound, freezable account. The spine: an agent stays sovereign while it stays crypto-native; the moment it needs real fiat, it hits the KYC wall.


The KYC wall: who holds the account?

Start here, because it determines everything else. KYC (“know your customer”) binds a financial account to a verified legal identity. An agent has no legal identity to verify — so it cannot open a KYC’d account itself. Two consequences fall out, and they split the whole field:

So the practical rule for a builder: an agent can remain fully sovereign as long as it stays crypto-native — holding Bitcoin, swapping to a stablecoin and back via non-custodial services when it needs a stable unit of account. The instant the workflow requires real fiat — a bank payment, a fiat invoice, a payroll deposit — the agent must borrow its owner’s delegated, KYC’d account. The KYC wall is where the parallel economy ends and the agent starts operating as its principal’s proxy.


What an agent actually does at an exchange

For a non-custodial swap, the flow is short: the agent calls the swap service’s API directly from its own wallet — no account, no delegation — and the swap settles atomically to its keys.

For a custodial venue, the pattern is longer and nearly identical across venues:

  1. Owner opens and KYCs an account — the compliance boundary; identity attaches here, once.
  2. Owner delegates to the agent — scoped API keys (ideally least-privilege: trade + withdraw-to-allowlisted-address only).
  3. Fund — from a bank (slow fiat rails: ACH, wire, SEPA), card, or by receiving BTC/Lightning.
  4. Convert — fiat↔BTC or BTC↔stablecoin via the venue’s API.
  5. Withdraw to self-custody promptly — bounds the custodial freeze surface to the time funds sit on the venue.

Two features make or break a venue for agent use. API access is non-negotiable — without it the venue is human-only. Lightning support matters more than it looks: a venue that pays out over Lightning lets the agent move funds off the venue in seconds for a fraction of a cent, shrinking the custodial-freeze window that an on-chain-only or bank-only venue leaves open.


Non-custodial, no-KYC swaps (agent-sovereign, crypto-native)

The agent swaps on its own keys — no account, no delegated identity. This is the most agent-native path; the caveats below are the price of that sovereignty. (Structural facts WebSearch-verified 2026-06-03; Boltz re-verified 2026-06-05. ✅ yes · — no · ⚠ limited; the API column is the capability an agent needs to run a swap unattended.)

ServiceTypeLightningStablecoin (network)APIBank fiat
Boltzatomic swapUSDT0 + USDC (via Circle CCTP: ETH/Arbitrum/Base/Polygon)✅ REST / boltzd
SideSwapLiquid swap(Liquid)L-USDt (Liquid)
SideShiftswapUSDT (Liquid) + 200+ assets✅ REST

Boltz is the standout for agents — no-KYC, Lightning-native, fully non-custodial atomic swaps, a REST API + boltzd for automation, Bitcoin across L1/Lightning/Liquid/Rootstock, and both major stablecoins (USDT0, plus native USDC live via Circle’s CCTP since May 2026). SideSwap is pure atomic swaps on Liquid (liquidity tracks order-book depth). SideShift spans the most assets but is not as clean as the other two: an automated risk-screening layer can flag and hold funds and may demand KYC/source-of-funds to release. None reach bank fiat — and that dividing line is the point: there is no no-KYC, API-driven, fiat-settling exchange, because that is exactly what KYC law exists to prevent.


Custodial venues (owner-delegated, KYC)

The regulated, centralized venues. The owner completes KYC and delegates the account to the agent by API key; the freeze surface is bounded by withdrawing to self-custody promptly. Two factual axes matter for an agent: what the account holds — a Bitcoin-only account confines the freeze surface to BTC, a multi-asset account exposes every asset held — and jurisdiction, which sets licensing, recourse, and availability.

VenueHoldsJurisdictionLightningStablecoin (network)API: dep / trade / withdrawBank fiat
StrikeBTC-onlyUS + ~95 countries✅ nativeUSDT (TRON, regional)✅ / ✅ / ✅
RiverBTC-onlyUS(RLS)✅ / ⚠ (RLS = Lightning payments, no buy/sell) / ✅
SwanBTC-onlyUS✅ / ⚠ (buy-only, DCA) / ✅
Krakenmulti-assetUSUSDC, USDT (multi-network)✅ / ✅ / ✅
Coinbasemulti-assetUSUSDC (Base/ETH)✅ / ✅ / ✅
Binancemulti-assetOffshore (global)USDT, USDC, FDUSD✅ fullrestricted (.US separate)
OKXmulti-assetOffshore (Seychelles)USDT, USDC✅ fullrestricted
Bybitmulti-assetOffshore (Dubai)USDT, USDC✅ fullrestricted
Bitget / MEXC / KuCoinmulti-assetOffshore (Seychelles)USDT (+USDC)✅ fullrestricted

Only the venues with a full deposit/trade/withdraw API — Strike, Kraken, Coinbase, and the offshore giants — can run a fiat↔BTC treasury unattended. River’s public API (RLS) is Lightning payments, not buy/sell; Swan’s automates buying (DCA) + withdrawal, not two-way trading — both stay useful for their niches (River for Lightning payouts, Swan for scheduled accumulation) but neither does programmatic conversion. The large offshore exchanges — Binance, OKX, Bybit, Bitget, MEXC, KuCoin — are the same animal as the US multi-asset venues under a different jurisdiction: offshore domicile adds regulatory and recourse uncertainty (several have faced enforcement or market exits) on top of the account-level freeze surface, but they hold the deepest stablecoin-and-BTC liquidity (the pools described in The Stablecoin Landscape). Across all of them, bank fiat — the one thing the non-custodial swaps can’t reach — appears only here; withdraw to self-custody promptly and treat any on-venue balance as exposed. (Volumes, jurisdictional availability, and listings shift constantly — see Field-Notes.)


Compliance lives at the gateway

For the custodial venues, the principle the Marketplace overview states holds: the venue runs its full regime on the account and the fiat leg; the BTC/Lightning leg downstream, once withdrawn to self-custody, is unrestricted. The pattern breaks when KYC is pushed into the protocol, when terms compel repatriation of self-custodied BTC on demand, or when every venue an agent uses terminates in one jurisdiction — answered by multiple independent venues across non-correlated regimes, prompt withdrawal, and hot/cold separation.


Two things that are not exchange


What good exchange infrastructure looks like for an agent

Reading down the comparison, a profile emerges for the venue best suited to an autonomous agent — the one that asks the least of a human and surrenders the least sovereignty:

The honest catch is that the last criterion fights the first four. The deepest liquidity lives on the large custodial, KYC’d venues; the purest sovereignty lives on the non-custodial swaps, which are thinner. So the “ideal” is a frontier, not a single winner — and today the venue sitting closest to it is Boltz: no KYC, non-custodial and atomic, a REST API, Bitcoin across L1/Lightning/Liquid/Rootstock, and both major stablecoins — with liquidity-at-size the one axis where the custodial giants still lead.

The standing build opportunity — scarcely filled — is a regulated agent-payment gateway on Lightning-substrate rails: the compliance assurances institutions need without compromising the Bitcoin leg. And the deeper open frontier, the one that would dismantle the KYC wall itself: agent-native identity — reputation systems, on-chain attestations, and zero-knowledge proofs that could one day satisfy a regulatory regime without a human’s delegated KYC. Until that exists, the wall stands, and the delegation pattern is the practical reality.