oniswap (@oni_swap) header
oniswap (@oni_swap) profile picture

The prop AMM for everyone else on Solana. FSfZuwaRFVcTtVBsiJ45NYK7Mwi3ANRcVT8kpbevtPGj

Joined November 2026
Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)Photo by oniswap (@oni_swap)
The code that prices a prop AMM is private to the trading firm that runs it. The code that prices an OniSwap pool is one file of 201 lines that anyone can read in a few minutes and run for their own pool. The OniSwap program uses its price only inside a band around Pyth, pauses swaps if it goes quiet, and gives its key no way to touch pool funds. oniswap.app/keeper.mjs
8
1
27
998
We took one day of JTO's real trading volume on Solana and replayed it against a $250,000 OniSwap pool, sending every trade to whichever paid the trader more, our pool or Jupiter's best route. The OniSwap pool won 76% of the trades and earned its LPs 60.7% a year, four times what an ordinary pool of the same size made. Every token we tested is in Table 6 of the whitepaper. oniswap.app/whitepaper
5
14
1,003
OniSwap now sends your swap to Jupiter whenever Jupiter pays you more than our own pools do. On oniswap.app you can pick any token Jupiter trades. The swap card prices our pools and Jupiter's best route for the same amount, takes whichever leaves more in your wallet, and adds no fee on top of the Jupiter route. Our pools get your trade only when they pay you more. oniswap.app/app
1
1
16
893
Uniswap's VolatilePair hook charges a temporary fee on the trade that pushes the price back after a big move, and only Uniswap Labs can open those pools. That half already runs on Solana. OniSwap's Backrun premium hook is live on mainnet, makes the trade that reverses a large one pay extra, and the pool's LPs keep it. Any wallet can open a pool with it. OniSwap pools can also follow a live price feed, so the quote has already moved before the first bot shows up. The other half, a fee that follows how fast the price is moving, is what we're building now. oniswap.app
VolatilePair Hook is live on Ethereum mainnet with ETH/USDC Instead of one static fee, the hook sets a fee for each side of the trade and shifts both with the market The result: LPs keep more of the value they create
1
13
663
A SOL/USDC pool on Solana earned $49,048 in swap fees in the last 24 hours. On OniSwap, 5% of every fee goes to the person who runs the pool, so a pool like that would pay whoever runs it $2,452 a day. You put your own copy of our curator hook on Solana, register it once, and open pools on it. From then on the OniSwap program sets aside 5% of every swap fee in those pools for your wallet, while the LPs keep 85%. Every step, from the kit to your first pool, is at oniswap.app/app#curators
12
2
29
1,255
For 30 straight minutes, the best SOL price on Solana came from pools nobody else is allowed to deposit into. We logged every venue's SOL/USDC quote once a minute. HumidiFi, ZeroFi, BisonFi and GoonFi held the top spot every single minute, while Orca, Raydium and Meteora sat 5 to 10 bp behind. Those four are prop AMMs, repricing with the market on their owners' money, and the fees they earn stay with their owners. OniSwap pools reprice from a live feed the same way, and the fees go to whoever deposits, from any wallet. oniswap.app
6
3
20
1,049
OniSwap pools can now hold 16 times the depth of an ordinary pool with the same money, right at the market price. That changes what it takes to win a trade. Against Jupiter's best live route for JTO, our first curve needed $710,000 in the pool to give traders more at every size from $1,000 to $50,000. The new deepest curve does it with $180,000. Whoever opens a pool picks its curve, from 1/2 up to 15/16, and the app shows it to every LP before they deposit. oniswap.app
3
2
15
1,089
On Solana today, the price to buy $10,000 of SOL and the price to sell it are within $3 of each other. For JTO the gap is $281, and PYTH and BONK are over $100 each. SOL gets that from prop AMMs like ZeroFi and BisonFi, which move their quotes with the market. Jupiter even uses one for JTO, turning your USDC into SOL on ZeroFi first, and almost all of the $281 comes from the next leg, SOL into JTO through public pools. OniSwap pools can follow a live price feed too, and the JTO/USDC feed is on mainnet now, so a JTO pool on OniSwap prices off the market instead of its own last trade. oniswap.app/app
7
2
23
1,340
We built a hook to steal from our own pool and gave it a full swap to try. In a pool that hands its hook the vault and signs for it, the hook left with half the money in one trade. In OniSwap the same hook left with nothing, and the vault finished the swap holding every token it started with. An OniSwap hook only ever gets its own memory account, so the vault is out of its reach and the pool keeps its signature to itself. That is why you can LP into a pool running somebody else's hook. The robber still lives in our test suite, and it has yet to take a single token. oniswap.app
3
2
15
827
Anyone can now run a prop AMM on Solana. A prop AMM moves its price with the market before bots can trade against a stale quote, and so far every one has run on its owner's own money. On OniSwap you pick a pair, create a price feed with one wallet approval, start our keeper with two commands, and open a pool on it that any wallet can LP into. The keeper is a single file that costs about 0.02 SOL a day in fees. The program uses its price only while it sits inside a band around Pyth, and the keeper's key can never touch pool funds, so your LPs are protected by the code itself. oniswap.app/app#feeds
23
5
34
2,239
Every big swap on a Solana pool leaves a gap, and a bot closes it a moment later with the LPs' money. On OniSwap, that bot now pays the LPs for it. The Backrun premium hook charges the trade that reverses a large one an extra fee that halves about every second, and the fee stays in the pool. Bots waiting for it to fall are racing each other, so the first one to move hands most of the gap back. Ordinary swaps rarely trip it, and about 9 in 10 paid the normal fee in our test. We ran a week of real SOL prices through a $1M pool, and the bots gave up more than half of what they used to take. Because the hook reads only the trades, any token pair can use it, and it is live on mainnet today. oniswap.app/app
9
3
25
1,790
We tested the worst day an OniSwap pool can have, a keeper in an attacker's hands that posts whichever edge of the price band pays the attacker most, every second. Over four weeks of SOL, the LPs in that pool still earned 34.9% a year, while an ordinary pool charging traders the same earned 12.9%. On all ten US stocks we ran, the attacked pool came out ahead of the ordinary one. On SOL and on every stock we tested, an OniSwap pool's worst case beat an ordinary pool's normal day. oniswap.app/whitepaper
4
1
17
985
We built a Solana pool that bots take 87% less from. Every time SOL moves, an ordinary pool is still quoting the old price when the first bot shows up, and its LPs pay the difference. We replayed four weeks of SOL, one price a second, through an ordinary pool and an OniSwap pool at the same 0.1% fee. The OniSwap pool followed the market off a keeper posting the real price, and the bots walked away with 87% less. Market makers have run pools like this on their own money for two years. OniSwap opens one to any wallet. oniswap.app/whitepaper
8
2
25
1,285
In April, @jump_ wrote that "the only barrier to PropAMM entry is writing better code and quoting a better price." Their own numbers back the pricing half, with the median SOL/USDC prop AMM fill landing 0.72 basis points from the best exchange midpoint across 20 million fills. There is a second barrier the line leaves out. A prop AMM quotes from its operator's own inventory, so the firm that writes the code is also the only one who can fund it. OniSwap keeps the repricing and opens the deposit. Pools follow a keeper's price that is checked against Pyth on every swap, and any wallet can LP. oniswap.app
1/ PropAMMs and the Next Chapter of Permissionless Market Structure For years, onchain trading won on access, lost on execution. Some feel better performance requires moving backwards to centralized systems PropAMMs on @solana are proving permissionless can outcompete centralized
2
2
11
1,000
0x went through 84,163 Uniswap v4 hooks and marked 54.2% of them malicious. Their worst examples quoted a clean price and took fees of up to 18% at settlement. An OniSwap hook can touch a trade in exactly two ways, the fee and a surcharge for its payee, and the program clips them at 10% and 0.5% whatever the hook asks for. The hook works only inside its own memory account, so the vault and the swap math stay with the pool, and the app shows each hook's program, its upgrade key and who it pays. Hooks were never the mistake, leaving their reach unbounded was. oniswap.app/app
A malicious hook doesn't need a UI to scam you. 👉 It just needs to look like the best quote. After analyzing over 84,000 v4 hooks, we determined only 19% of hooks to be safe. It's time to get real about hooks.
2
1
12
899
A hook on OniSwap costs about 2,000 compute units a call. We measured it on a local validator. A swap with no hook used 825 units, and the same swap calling its hook before and after used 4,772. Those two calls take 0.28% of the 1.4 million units a Solana transaction may spend. For that you get a rule of your own on every pool that picks your hook, a fee that follows the logic you write, and under today's config 5% of every fee those pools collect, paid to your hook's payee. The pool hands your hook its own memory account and nothing more, so the money stays with the pool. oniswap.app/app
1
2
17
859
The rule behind this ceiling is built into the button that opens a pool. When you open an OniSwap pool on a price feed, the app works out the lowest fee that still covers the feed's band after the protocol's share and the hook's share leave the pool, and it keeps every new pool at or above it. On the SOL/USDC feed, with its 0.2% band, that floor is 0.23% for a plain pool and 0.24% with the Launch fee hook. Half the band was the textbook answer. On OniSwap the fee covers all of it, and the app does that math for you. oniswap.app/app
How much can a lying keeper take from a $1,000,000 OniSwap pool? $13.78, once. That is the worst case with a 20 basis point band, the keeper lying as far as the band allows, Pyth itself running 50 basis points late, and trades of up to a thousand times the pool's reserves. When Pyth is current the figure is $1.12, and a pool sitting at the market price gives up nothing at all. Those numbers come from an exact search over every sequence of trades, and they land within 0.25% of the bound in our proof. The ceiling holds because the fee a pool keeps covers the whole band. With the textbook fee of half the band, the same attacker drew about $21,000 out of the same pool in twelve round trips and was still gaining on the last one. oniswap.app/whitepaper
1
4
16
936
In September, Ozan Solmaz, Lioba Heimbach and @jason_of_cs published the first full year of data on prop AMMs. On Solana, two seconds after a fill, prop AMMs earned 0.37 basis points while ordinary AMMs lost 0.22. They traced that edge to four habits. Three are good engineering, repricing without waiting for a trade, pricing the risk each counterparty brings, and staying clear of arbitrage from other venues. The fourth is filling trades at a worse price than the one quoted. OniSwap keeps the first and has no room for the last. Its pools reprice from the market the way a prop AMM does, every fill comes from a public curve that anyone can check, and the fees go to every wallet that deposits. arxiv.org/pdf/2609.38056 oniswap.app
3
4
17
1,221
On OniSwap, nothing stands between an LP and a withdrawal. Deposits and withdrawals run on the pool's reserves alone, without its hook and without its price feed. A hook can stall swaps in its own pool, a keeper can go quiet and Pyth can fall behind, and through all of it every LP can still take their share out. We tested exactly that with a hook that fails on every call, and the full withdrawal went through. The program also has no switch to pause or freeze a pool, and that includes ours. oniswap.app/app
2
5
14
953