01
Money that
moves like
an API call.
Cross-chain stablecoin settlement for agents and humans. Gasless, multi-route, one call.
ETHGlobal Online 2026
6 chains
SDK · MCP · x402
02
THE PROBLEM
Agents can think. They can't pay.
An AI agent that owes another agent $5 for an API call has no way to settle it across chains. Neither does a person holding USDC on the wrong network.
01
Gas is a dead end
You need the native token of a chain you've never touched, before you can move the stablecoin you already hold.
02
Bridges assume a browser
Connect wallet, approve, sign, wait. There is no programmatic path for software that has no hands.
03
Bridges assume you pay yourself
Most move funds between your own addresses. Paying someone else on another chain is a different product.
03
THE PRODUCT
A settlement rail with four front doors.
One signature. No gas. Funds land with anyone the payer names, on any supported chain.
STEP 01
Sign
Payer signs an EIP-3009 permit. A signature, not a transaction.
STEP 02
Relay
Breeja submits the deposit on the source chain and pays the gas.
STEP 03
Route
Router scores live gas, liquidity and preference, then picks a path.
STEP 04
Settle
Funds released to the named recipient on the destination chain.
Neither payer nor recipient ever holds a gas token.
04
SAME RAIL, FOUR CALLERS
Built for software first.
Human · Web app
Connect a wallet, pick chains, sign, watch it land live.
Agent · SDK
One call, no browser, no gas.
await breeja.pay({ … })
Agent · MCP
Claude pays across chains as a tool call, with spend caps enforced.
Agent · x402
Hit a paywalled endpoint, settle, retry with proof. Cross-chain.
All four consume the same SDK. The SDK is the product — the web app is just one client.
05
DEVELOPER EXPERIENCE
The whole integration.
import { Breeja } from "@breeja/sdk";
const payment = await breeja.pay({
from: "base-sepolia",
to: "arbitrum-sepolia",
amount: "10.00",
recipient: "alice.eth",
signer: account,
});
// → settled in seconds, no gas token held
Runtime discovery
Agents call chains() — never a hardcoded list.
Typed errors
Stable codes an agent can branch on exhaustively.
Idempotent
Keyed on the permit nonce. Retries never double-spend.
06
ROUTING
A real choice, not a single path.
Every payment is quoted across routes and scored on live gas, pool liquidity, and caller preference.
Route
Latency
Cost
Custody
Best for
Fast pool
~10s
0.5%
Custodial
Small, latency-sensitive payments
CCTP
~15m
gas only
Trust-minimized
Large transfers, canonical guarantees
Callers override with preference: "fast" | "cheap" | "trustless". Custody risk scales with amount; latency tolerance doesn't.
07
THE AI LAYER
Models explain. They never decide.
The boundary is enforced in code, and it's why the system is auditable.
LLM permitted
Parse natural-language intent
Explain a completed decision
Deterministic
Validate the request
Select the route
Compute the fee
LLM forbidden
Choosing where money goes
Authorizing a release
Every number in a generated explanation is validated against the decision object. Mismatch falls back to a template — a model never states a fee the system didn't compute.
08
COVERAGE
Every chain is both a source and a destination.
Base
Sepolia · 84532
Fast pool · CCTP
Arbitrum
Sepolia · 421614
Fast pool · CCTP
Optimism
Sepolia · 11155420
Fast pool · CCTP
Arc
Circle's L1 · native USDC
Fast pool · CCTP
Hedera
Testnet · 296
Fast pool
Ethereum
Sepolia · 11155111
Source only
A payment is a directed edge between any two. Not a hub-and-spoke bridge.
09
TRUST MODEL
Stated plainly, not buried.
WHAT WE MITIGATE
—
CCTP offered as a trust-minimized alternative for the same payment
—
Pool ownership held by a Safe multisig, not a single EOA
—
Release authority is Ledger-backed, not a hot server key
—
State is durable and reconciled from chain events
WHAT WE DON'T CLAIM
—
The fast-pool route is custodial — the relayer controls liquidity and decides when to release
—
Release is not decentralized. A bonded watcher network is the next real trust reduction, and it is not in this build
Same tradeoff early Across and Hop shipped — and the honest reason the fast route is fast.
10
INTEGRATIONS
Each one load-bearing.
Circle / Arc
CCTP settlement route + native USDC chain
The Graph
Indexed payment history across every chain
Privy
Embedded wallets and agent server wallets
ENS
Agent identity — pay a name, not a hex string
Ledger
Hardware-backed release authority
Hedera
Source and destination chain
Bazantic
Agent-ready SDK and MCP tooling
World
Verified human authorizes an agent's spend cap
Deliberately skipped: AMM and DeFi-position tracks. A swap bolted onto a payment rail is the integration judges see through.
11
STATUS
Shipped, in flight, next.
EIP-3009 gasless permits, live on real testnets
SHIPPED
Contracts covered by tests, deployed and verified
SHIPPED
Agent-to-agent payment, no browser in the loop
SHIPPED
Durable state, idempotency, reconciliation
IN FLIGHT
Multi-chain mesh + CCTP route scoring
IN FLIGHT
SDK, MCP server, x402 demo
IN FLIGHT
Bonded watcher network — decentralized release
NEXT
Mainnet with funded liquidity and multisig custody
NEXT
12
THE ASK
Agents are about to need
a way to pay each other.
Breeja is that rail — and it works for people too, through the same contracts.
TRY IT
Install the SDK and settle a cross-chain payment in one call.
PLUG IN
Add the MCP server and let your agent pay across chains today.
BUILD ON IT
Use Breeja as the settlement leg behind your own x402 endpoint.