One catalog for AI compute, supplied by the browsers that use it.
Apps and people buy small inference jobs from the OCLO catalog. Open tabs run them, every result is checked, and only verified work is settled on Solana, paid from what the buyer put in.
Click an open tile to watch one job go from claim to check to payout. Keyboard: Tab into the mosaic, arrows to move, Enter to start.Tap an open tile to watch one job go from claim to check to payout.
Pay comes from a buyer's job, never from the clock.
There is no reward for keeping a tab open. A job pays when a buyer funded it, a device ran it, and the check passed. The money a contributor receives is the money a customer paid for that work.
The count on the tab is jobs that passed the check. It moves only when a buyer's job is done.
One job, from quote to settlement
Quote. The buyer picks a job from the catalog and sees the price before anything runs.
Assign. The scheduler cuts the batch into independent jobs small enough for a browser.
Run. A light public model runs on the contributor's device, inside the time and resource caps they set.
Check. Each result is compared against a reference before it joins the batch. In the first release the operator runs these checks, in the open.
Settle. Verified work settles on Solana once. A job cannot be paid twice.
Refund. Whatever was not delivered goes back to the buyer by rule.
- 1. Quote
The buyer picks a job from the catalog and sees the price before anything runs.
At the tile: The price is set before any device moves.
- 2. Assign
The scheduler cuts the batch into independent jobs small enough for a browser.
At the tile: The scheduler hands this job to Tab A, inside the caps its owner set.
- 3. Run
A light public model runs on the contributor's device, inside the time and resource caps they set.
At the tile: Running in this tab. Only the result is sent back.
- 4. Check
Each result is compared against a reference before it joins the batch. In the first release the operator runs these checks, in the open.
At the tile: Compared with a reference result. No match, no stamp, no pay.
- 5. Settle
Verified work settles on Solana once. A job cannot be paid twice.
At the tile: Paid once, to the device that did the work.
- 6. Refund
Whatever was not delivered goes back to the buyer by rule.
At the tile: Work no device delivered goes back to the buyer.
Contribute this tab
See what this browser can run, say how long and how hard you would like it to work, and take one job. Inputs are public images, and you can stop at any time.
- WebGPUNot checked yet
- WebAssemblyNot checked yet
- Threads this browser reportsNot checked yet
Mallard duck, Thomas G. Barnes/U.S. Fish and Wildlife Service, Public domain
Your preference: up to 30 min and about 50% of the device.
When a device leaves, its jobs move to another
Tabs get closed and phones go to sleep. The scheduler watches every assignment and hands unfinished work to the next device. Only the device that finishes and passes the check is paid.
Three devices, each holding its own jobs.
$OCLO holds a reserved slot
Jobs are paid in USDC, and running a batch needs no $OCLO. The token is for capacity that is scarce: a slot in a period when real supply exists. A repeat buyer will lock $OCLO to hold the slot and prepay its jobs in USDC. The lock holds the place; the jobs run at their quoted price.
From launch, holding $OCLO lets a wallet request reserved capacity. Two locks open with the reservation program: a slot lock, and a standing lock that gives a buyer's jobs priority when work is assigned. Both come back when released, and if supply fails the slot is released with its USDC.
Half of every creator fee from $OCLO trading, plus 25% of the service fee income, buys $OCLO on the market and burns it. It runs from launch: half of each creator fee as it is claimed, the service share every 24 hours, and every buy and every burn is a transaction on chain.
Pick a week with supply to hold a slot.
Held here: 1Slots held here stay on this page.
One catalog for every app that needs a model run
We started with image batches because they are small, independent and easy to check. We are building toward the next question: what happens when the apps you already use buy a classification or a search from this same catalog, and the supply behind it is people paid per verified job?
Image classification
Public image batches, a light public model, each result checked against a reference.
First releaseSemantic search
Embed a batch once, then search it by meaning.
First releaseMore job types
A job type joins the catalog when its results can be compared against a reference.
Added when a buyer needs them and they can be checked- People join with a tab and run batches
- Repeat buyers reserve real capacity
- Apps buy compute through the same job catalog
Every payment has a row
Each settlement is one row: the job, the device that ran it, the check, the amount and the transaction. Refunds get rows too.
The jobs this page ran, as rows. Amounts and transactions are filled in when real work settles.