Robinhood Chain Testnet$EBBAI credit every 30 min
Hold $EBB and the fees from every trade pay for your AI, landing as a fresh tidepool of credit every thirty minutes. Credit you leave on the shore for seven days is drawn into the Trench: it buys $EBB back and burns it.
Fee-to-credit tokens proved that people want their AI paid for by trading fees. Then the undertow showed up: most credit is never spent, and where that unused value flows decides whether holders are being paid or quietly drained.
of credit issued by an incumbent fee-to-credit token was never spent on inference. It was listed for sale below face value.
As reported on orbz.app (October 2026), citing an incumbent's public analytics. Not independently verified by us.
Seats, keys, top-ups, overages. For anyone who builds with models the meter only runs one way.
Holding a token rarely changes what you do on an ordinary Tuesday. A key that answers your editor does.
Pay unused credit out as a tradable token and holders sell it at a discount. You get paid in something you dump.
Stake, mint, list, settle, activate. Most holders stop at the second step and their share sits idle.
Eight stations, one loop. Nothing in it waits for a person to press a button, and every station leaves a public trace. Scroll to send a drop of fee around the current.
Every buy and sell pays 3% on Pons: its 1% base fee plus the 2% $EBB creator tax, on the bonding curve and after graduation alike. Pons keeps 0.3%, so 2.7% of all volume belongs to the Basin, in ETH.
Pons fee escrow → EbbVaultPons does not push fees, so anyone can call harvest(). It claims the ETH from Pons’ fee escrow and swaps it to USDG on the WETH/USDG pool, reverting if the price strays more than 3% from the 30-minute TWAP.
harvest()The swap result is split in the same transaction: 70% is booked to the current tide’s credit pool, 30% goes to the treasury. An invariant test checks the ratio on every harvest.
toPool × 3000 == toTreasury × 7000When a tide ends the allocator integrates every wallet’s balance over those 30 minutes. A time-weighted 100,000 $EBB or more is eligible; flash holds earn almost nothing.
TWAB over [start, end)Each holder’s pro-rata grant is committed on-chain as a Merkle root, and lands in a fresh tidepool in their dashboard. Anyone can check their own leaf against the root.
commitGrants(tide, root, total, wallets)Point any OpenAI-compatible client at the gateway. Each request reserves its maximum cost from your oldest pool first, then settles the real cost and refunds the rest.
POST /v1/chat/completionsA tidepool lives 336 tides, seven days. Spent credit is paid to the model provider hourly, with a Merkle root of the requests it covers, and never more than the tide was granted.
withdrawForUsage(tide, amount, usageRoot)At hour 168 the remainder buys $EBB on the market and burns it. Our keeper calls it; if the keeper stops, anyone can, and collects a 0.25% tip capped at $2.
burnExpired(tide, maxAmount)Spend a credit and it becomes AI. Leave it, and at hour 168 it becomes a buyback. Scroll, or drag the brass handle, to follow one dollar of credit through its week.
Hour 0, day 1. 168 hours until the Trench.
FIFO. Requests draw on the pool closest to the Trench, so you never lose fresh credit to an old countdown.
Credit belongs to the wallet the grant was committed to. Keys spend it; they cannot move it.
It buys model usage at list price. It never turns into ETH, USDG or anything else.
An expired pool is already buying $EBB. It does not come back.
Turn a dial and the strip reprints. Every line comes from the fixed fee split in the contract, not from a projection. Switch to the latest tide to read a real entry from the Logbook.
pool, curve, vault, treasury and burn address never count
A model, not a forecast. Grants follow real volume tide by tide and are never promised.
Every burn takes the same public path: USDG to ETH to $EBB, then burn(). Total supply falls by exactly the amount burned, and each burn prints a Logbook entry.
// anyone, once the tide is 336 tides (7 days) oldEbbVault.burnExpired(tide, maxAmount) require(now >= epochStart(tide) + 336 * 1800) usdg = min(remaining(tide), maxAmount) // slices tip = min(usdg * 0.25%, $2) // to caller eth = swap(usdg - tip, USDG -> ETH) // 3% TWAP bound ebb = swap(eth, ETH -> $EBB) // 3% TWAP bound EBB.burn(ebb) // supply falls emit Burned(tide, usdg, ebb, msg.sender, tip)Whatever is left in a tide after seven days. The main pipe; it runs every tide.
Credit bought at list price plus 5%. The 5% goes down the same pipe.
The treasury’s free-demo allowance nobody used, routed into the Trench.
Each burn is a Logbook entry with its transaction, the caller and their tip.
No staking, no locking, no second token. Depth buys room, not priority: every credit is worth the same dollar of usage, and grants are strictly pro-rata to time-weighted balance.
Tiers raise rate limits and add cosmetic marks. They never change what a credit is worth or how a tide is split.
Trading fees are the only engine at launch. Slide the spend share to see the credit pool divide between AI and the Trench.
At any spend share, the whole credit pool goes back to holders: as AI they used, or as a burn that shrinks supply.
Fig. 9 Published monthly in Soundings, with a transaction link for every movement.
The chain holds the money and the rules. Our servers meter, allocate and publish, and every step they take leaves a trace you can check against the chain.
2.7% of every trade, as ETH. Fees collect on the curve or Pons’ hook, are swept here, and wait until harvest() claims them.
Books USDG per tide, commits grant roots, pays settlement, burns expired credit. No owner, no upgrade path.
Where trades happen, and where every burn buys before calling burn().
Turns each tide into a Logbook row and a receipt card, and serves Soundings.
Hashes keys, rate-limits by depth, reserves cost from the oldest pool, settles the real cost.
Hourly, pays providers for used credit via withdrawForUsage, with a Merkle root of the requests.
Calls harvest() and burnExpired(). Its key holds only gas, and both calls are open to anyone.
At :00 and :30, computes time-weighted balances and commits the grant root.
Reads every $EBB transfer, three blocks behind the head, into balance timelines.
Sign one message, paste a key into any OpenAI-compatible client.
Named upstream models at list price. More than one, so no single provider is a dependency.
Can call harvest() or burnExpired() and earn the tip. Nobody can stall a burn.
base_url = "https://api.ebbtide.xyz/v1"api_key = "sk-ebb-…" # one signature, no transactionThe Basin is tested against eight invariants under randomised call sequences. Watch the beam sweep the compartments: if one floods, the others keep the hull afloat.
USDG in the Basin equals the sum of every open tide’s remainder
usdg.balanceOf(vault) == Σ remaining(e)A tide never pays out more than it granted, nor grants more than it booked
withdrawn(e) ≤ granted(e) ≤ booked(e)Nothing can be burned before the tide is seven days old
burnExpired reverts before epochStart + 7dA burned tide can never be withdrawn again
withdrawForUsage(burned) revertsEvery harvest splits exactly 70 / 30
toPool × 3000 == toTreasury × 7000Swaps revert beyond 3% of the 30-minute TWAP
|out − quote| ≤ 3%Total supply drops by exactly the amount burned
ΔtotalSupply == ebbBurnedUSDG leaves only to settlement, treasury, the burn path or the caller tip
no other transfer target exists
Every request to /api/soundings reads the Basin's USDG balance from the chain and sets it against everything the books say is in it. The equation is the whole proof of reserves; nothing is taken on trust.
The product ships before the token. Each waypoint is reached when its work is public and checkable, not when a date arrives.
Fig. 12 The charted course. Solid line is sailed; dashes are still ahead.
Launch and the first burn are the same mark on the dial, seven days apart. Everything in between prints to the Logbook.
The whole idea, in one line
Everything you spend becomes AI. Everything you leave becomes a burn.
Fair launch · 5 Oct 2026 · 15:00 UTC · @ebbtidexyz · Telegram