Use Cases
Market Makers
Hedge RWA exposure through one API. Atomic fills, settled in stablecoins.
A quoting desk that lists RWA perpetuals takes inventory it then has to offset. Gateway is that offset: gold, FX, indices, commodities, ETFs, and single stocks behind one API, settled in the stablecoins you already hold. You send a signed intent. You get a fill or a decoded reject. There is no transaction to wait on, and no partial state to carry on the book.
Gross exposure, netted
One integration
Prices, orders, positions, depth, and a streaming feed share one surface. A book that quotes gold, NDX, and EURUSD does not grow a venue per underlying.
Hedging, netted
When the book needs the other side, you send a signed intent. A per-size executable quote lets you see the fill before you send. Gross can grow both ways. Net is the hedge you chose.
Stablecoin settlement
Onboarding relays a USDC permit and delegation in one call. Trading does not need ETH. Proceeds settle to the trader's address. A stolen relay cannot redirect funds.
Working with the book
Quoting
Your quotes stay on your venue. Gateway is the reference and the hedge. The production feed ticks at about 465 milliseconds median across monitored pairs, with a snapshot on connect. Over a three-minute tape the largest gap observed was 1.8 seconds. The stream is simple enough to cache on the client, which is how a desk decides on a print it can still see.
Before you send, /v1/depth/quote returns an executable price for the size you will actually trade, not the inside of a book you will never lift. A synthetic ladder sits under that quote. Against a recorded tick tape, fifty tick-triggered round trips predicted the fill to 0 bps. That is the difference between hedging the print you saw and hoping the next one is close.
Eighty-three instruments sit behind the same five endpoints: crypto, FX, indices, commodities, ETFs, and single stocks. A book that is long gold from a client quote and short NDX from another does not grow a second vendor to flatten both.
Offsetting
Putting on a clip and flattening one are the same call. Every field of the intent is signed. The delegate contract will not execute an order that does not carry that signature. Short deadlines and one-shot random nonces are the rest of the model, which is why recovery is a decoded reason rather than a hash in a mempool.
One HTTP POST returns the terminal outcome: a tradeId, fill price, and amounts, or a revert with a reason. No gas, no polling, no partial state. Across about 1,900 live production orders that path never came back ambiguous. Wall time, POST /v1/orders to terminal, was about 0.7 seconds on warm connections with websocket-cached pricing.
Twelve days of aggressive automated testing, including two stage incidents traded through on purpose, closed every position that opened. That is the cleanliness a desk needs when the hedge sits on someone else's venue rather than its own matching engine.
Margin and reporting
Margin treatment is yours to set on top of the fill. What the API returns is a terminal state: a fill, a tradeId, or a human-readable error. Production rejects come back fully decoded. You record that. You do not reconcile a transaction that might still be in flight.
Onboarding relays a USDC permit and a delegation in one call. After that, trading needs no ETH. Proceeds settle to the trader's address, so a compromised relay cannot redirect funds. Observed cost on production validation was about ten cents a round trip at $5 test size, oracle included, about $19 across 200+ real orders.
Those are test-size economics, from live orders, not a simulator. Latency is the full wall time of POST /v1/orders to terminal, over warm connections. They are not a guarantee at larger size. Latency found in stage on a Monday was measurably better the same day and resolved by Thursday, a 60% median improvement under load, verified on re-runs.
Access global markets. Start in minutes
Enterprise access for exchanges, market makers, stablecoin funds, and professional desks. Onboard with a solutions engineer.