PERMANENTLY FUNDED COMPUTE
Software that
outlives its owners.
corpusAI capitalizes an on-chain endowment for a keeper, an agent, or any statically deployed service. The principal is never spent. The yield pays the compute bill, month after month, with nobody in the loop.
Use cases
What is worth endowing.
Anything with a small, predictable, indefinite bill and no natural owner to keep paying it.
Keepers and liquidation bots
Critical to a protocol, funded out of somebody’s discretionary budget, and quietly the first thing cut.
Autonomous agents
An agent with a monthly inference budget becomes a permanent participant instead of a demo with a runway.
RPC nodes and indexers
Public read infrastructure that everyone depends on and nobody has a business model for.
Oracles and price feeds
Feeds whose value comes entirely from never stopping, which is exactly what recurring funding cannot promise.
Archives and static sites
Documentation, datasets, and records that should outlive the person who published them.
Public-goods infrastructure
Anything currently living grant cycle to grant cycle, converted into a position that does not need renewing.
Mechanism
How an endowment funds compute.
Four steps, one of which happens once and three of which happen forever.
Price
We model the workload’s compute cost and its variance against expected net yield, then size the principal with a buffer that absorbs both moving against you.
Fund
You pay the up-front price once. Inside the endowment it splits in two: principal, which is locked and can never be spent on compute, and a buffer, which can.
Earn
Principal sits in a blended yield-bearing reserve and accrues continuously. Yield is whatever the position is worth above the locked principal, which is what makes it measurable at any moment.
Pay
A harvester sweeps accrued yield above a threshold to the payout address that settles the compute bill. It can withdraw yield and nothing else.
The model
The whole product is one equation and a buffer.
A workload that costs C per year needs principal that throws off more than C per year, with enough headroom to survive a bad stretch of yield.
At k = 1 the coverage ratio is exactly 1.0, which means no headroom against a single bad quarter. Every figure quoted on this page uses k = 1.15.
| Monthly | 4% | 6% | 8% | 12% |
|---|---|---|---|---|
| $15 | $5,175 | $3,450 | $2,588 | $1,725 |
| $30 | $10,350 | $6,900 | $5,175 | $3,450 |
| $75 | $25,875 | $17,250 | $12,938 | $8,625 |
| $150 | $51,750 | $34,500 | $25,875 | $17,250 |
| $500 | $172,500 | $115,000 | $86,250 | $57,500 |
| $1,500 | $517,500 | $345,000 | $258,750 | $172,500 |
The problem
Nothing on the internet is paid for forever.
Every persistent service has a person behind it, and that person eventually stops paying.
- ✕A liquidation keeper stops the month a corporate card expires.
- ✕An indexer goes dark one cycle after the grant that funded it ends.
- ✕An agent runs exactly as long as someone remembers to top up its balance.
- ✕A public archive lasts precisely as long as its author stays interested.
- ✓The keeper's bill is paid by yield the month after everyone forgets it exists.
- ✓The indexer's funding has no cycle to end; principal never runs out because it is never spent.
- ✓The agent's balance tops itself up: a harvester sweeps yield to the payout address on schedule.
- ✓The archive outlives its author, its host contract readable by anyone who wants proof.
None of these are technical failures. The dependency is administrative. Every one of these services would have kept running indefinitely if the bill had simply kept getting paid.
FAQ
Questions worth asking.
A serious buyer will ask all of this before they email. It is better answered here.
Principal, current value, accrued yield, and every harvest are on-chain. You do not have to take our word for the balance.
The owner can change roles, the payout recipient, and can withdraw principal in an emergency. We name this rather than soften it, because sophisticated readers assume it and trust the ones who say so. For a protocol funding its own keeper, that owner is normally the protocol’s multisig — corpusAI does not need to be the owner, and in most deployments should not be.
sUSDe (staked USDe, Ethena) is the core position, accruing through share-price appreciation rather than rebasing. Part of the reserve sits in tokenized money market funds backed by short-duration government paper, so the position is not wholly exposed to any one crypto yield source.
Harvested yield goes to a payout address that settles the bill — today, a crypto card wallet paying a cloud invoice. The contract does not care what sits at the payout address, which is what lets this cover inference credits, bare metal, or a hosting invoice equally well.
The buffer absorbs it first. Past that, coverage drops below 1.0 and the position needs a top-up. At the quoted k = 1.15, net yield has to fall by more than 13% before coverage is at risk.
The buffer is sized against price drift as well as yield variance. A structural repricing of the workload is a re-underwriting — adding principal by the same equation quoted above — and we would rather quote that than pretend otherwise.
The contract, the principal, and the payout address do not depend on us continuing to exist. Anyone can run the harvester, which is permissionless by design. An endowment whose survival depends on a company is not an endowment.
Get started
Fund something that does not need funding again.
Tell us the workload and its monthly bill. We will come back with a principal, a buffer, and the stressed case in writing.