APIKEY is the settlement and governance asset of apikey.help, a DeepSeek inference gateway that resells the same model at 94% below the official list price. The token is not a promise of future utility — it already sits at the end of a live payment path, on a product with paying customers, real upstream cost, and a measured gross margin.
0x00000…66666. Mined with CREATE2, verified twice.Selling model access is a thin-margin business dressed up as a software business. The reseller buys tokens from an upstream provider, marks them up, and hopes the spread covers payment fees, chargebacks, and idle capacity. Three things break this model in practice.
A reseller advertising "50% off official price" is usually comparing against a number nobody pays. Official DeepSeek pricing has peak and off-peak tiers, and cache-hit input is priced at a small fraction of cache-miss input. If your workload is repetitive — and most production workloads are — your real cost is far below list. Conversely, if your workload is short and unique, your real cost is close to list and the advertised discount is a loss. The spread on list price tells you almost nothing.
Every top-up creates a liability the operator owes in compute. Customers have no visibility into whether that liability is funded. When a reseller disappears, prepaid balances disappear with it. This is not a hypothetical: it is the normal failure mode of the category.
Even resellers that survive are single-operator businesses. Pricing, treasury, and payout decisions happen in a private spreadsheet. Customers and partners have no mechanism to verify or influence any of it.
apikey.help is a live OpenAI-compatible gateway serving a single model:
deepseek-v4.1-flash-1017. Customers point their
existing SDK at https://apikey.help/v1, change two
lines of configuration, and their code runs unchanged. Billing is per token,
with no monthly fee and no charge for failed calls.
| Line item | Ours (USD / 1M) | Official off-peak | Discount |
|---|---|---|---|
| Input · cache miss | $0.00946 | $0.150 | 93.7% |
| Output · incl. reasoning | $0.03752 | $0.600 | 93.7% |
The discount is not funded by burning capital. It is funded by the gap between list price and actual upstream cost, plus prompt caching. Measured on production traffic: upstream cache ratio 0.0244, observed cache hit rate 95.4%, which produces an overall gross margin of 94.3%. The list-price spread alone would be only 4–5% — the margin comes from caching, and it is real, but it is also workload-dependent. A customer sending short, unique prompts every time gets a much thinner margin, and that scenario is disclosed in the risk section rather than hidden.
That last point matters for the token: the payment rails, the accounting, and the customer base already exist. APIKEY is being attached to a working system, not used to fund the construction of one.
APIKEY is a standard ERC-20 with three extensions that each remove a class
of trust: ERC20Burnable for verifiable supply
reduction, ERC20Permit so approvals can be signed
off-chain, and ERC20Votes so governance weight is
snapshot-based rather than flash-loanable.
| Property | Value | Verifiable how |
|---|---|---|
| Name / symbol | apikey / APIKEY | On-chain name(), symbol() |
| Total supply | 1,000,000,000,000 | totalSupply() — fixed at construction |
| Decimals | 18 | decimals() |
| Mint authority | none | No _mint outside the constructor; isMintable() returns false |
| Owner | none | Contract has no owner() function at all |
| Pause / freeze | none | No pause, no allowlist, no blacklist |
| Anchor price | $0.001 | priceAnchor() — display only, never used in arithmetic |
The founder address is a compile-time constant, not
msg.sender. This is deliberate: it means the
deployment transaction's sender cannot influence where supply lands, so the
recipient is auditable from source before the contract is ever deployed.
60% of supply is never controlled by a person. It sits in the treasury contract, which has no owner and no function that lets anyone withdraw directly. Every outflow requires a token-holder vote, a two-day timelock, and is capped at 30 billion tokens per transaction — 3% of total supply — so a single compromised key cannot drain it in one move.
20% is locked by code, not by promise. The vesting contract releases linearly per second over 1080 days from its own deployment timestamp. There is no revoke switch, no owner, and the beneficiary is immutable. The founder cannot accelerate it, and cannot recover the tokens by any path in the contract.
The remaining 20% is the founder's, unlocked, and disclosed as such. It is the portion that funds operations, upstream prepayment, and the sale inventory that governance later decides whether to deploy.
Every address below begins with five zeros and ends with five sixes:
0x00000…66666. This is not cosmetic. A predictable,
typo-resistant address family is a real operational security property: users
can verify a contract by glancing at its prefix and suffix, and a
clipboard-hijacking malware that swaps an address will produce something
that visibly fails the pattern.
The addresses were mined with CREATE2 against a fixed init code hash, then verified by two independent implementations before deployment. Each contract embeds the addresses of the contracts it depends on as constructor arguments appended to init code, so wiring the system together does not change any bytecode — and therefore does not invalidate any mined address.
0x000004124c09745a7c6f74584ea4e2ce5b866666
0x00000cd42f326eee20a8d567b4d849824ca66666
0x0000028c976f5c48f86d5fb053c4d3d21f666666
0x00000322ac269a7c6fee5ceed66c41dfd1066666
0x000009464589b9ad9cf1215e39df01bc21f66666
0x000000bb2e2634193a632092e7f53dd62c1666666
claim() every block.
This address is mined and verified but the contract is not yet on
chain — it is deployed only after the treasury has funded a reward
period, so no one can stake into an empty pool.
The token does not need to be the reason people pay. People pay because the inference is cheap and the endpoint works. The token sits downstream of that revenue and gives it somewhere to go.
| Flow | Mechanism | Effect on supply |
|---|---|---|
| Customer pays in USDC | Gateway records the order, backend credits the account | none |
| Customer pays in APIKEY | Valued at min(anchor, oracle), 20% burned on settlement | −20% of top-up |
| Primary sale | 20% of proceeds to treasury, equivalent tokens burned from inventory | deflationary |
| Protocol revenue | Accumulates in treasury; deployment decided by vote | none until spent |
| Staking rewards | Treasury-funded, never minted | redistributive |
A burn that requires a team decision is a burn that can be postponed forever. Both burn paths here execute inside the same transaction as the revenue event: the gateway burns on settlement, the sale contract burns at the moment of purchase. Nobody has to remember to do it, and nobody can decide not to.
Voting power is read from a past block, not the current one. This makes flash-loan governance attacks economically pointless: borrowing a trillion tokens for one block gives you zero votes, because the snapshot already happened.
Audits describe what was checked. This section describes what is structurally impossible, which is a stronger claim and one you can verify yourself by reading the deployed bytecode.
| Attack | Why it fails |
|---|---|
| Inflate supply | No mint function exists outside the constructor. Not "guarded" — absent. |
| Freeze a holder | No allowlist, blacklist, or pause. Transfers cannot be blocked by anyone. |
| Drain the treasury | No owner, no withdraw function. Only a passed proposal with a 2-day timelock, capped at 3% per transaction. |
| Accelerate founder unlock | Vesting has no revoke, no owner, immutable beneficiary, and time is read from the chain. |
| Flash-loan governance | Votes are read from a past block snapshot; borrowed tokens carry no weight. |
| Pump-and-redeem via oracle | Gateway values token top-ups at min(anchor, oracle). Raising the oracle cannot raise credit value. |
| Redirect sale inventory | Inventory withdrawal target is an immutable treasury address, not a settable field. |
| Erase staking rewards | USDC stream keeps a per-share remainder so block-level claims cannot round rewards to zero. |
Three contracts do have an owner, and it is the founder address: TopUpGateway and TokenSale can have their parameters adjusted (rates, caps, oracles), and APIKEYStaking will have an owner for reward notification and pausing. This is disclosed rather than papered over. The mitigation is that none of these owners can mint, cannot move user funds, and cannot touch the token or the treasury — the privileged surface is limited to parameters that governance can later take over via a proposal.
These contracts have not been audited by a third party. That is stated here rather than buried: treat unaudited code as a real risk regardless of how carefully it was written.
| Stage | Status | Scope |
|---|---|---|
| Gateway live | done | OpenAI-compatible endpoint, multi-provider failover, USDC billing, referral and withdrawal rails. |
| Contract suite | 5 of 6 deployed | Token, vesting, treasury, gateway and sale are on Base mainnet. Staking is mined and verified, awaiting treasury funding. |
| Treasury activation | next | Delegate votes, open the first governance proposal, publish the treasury address flow. |
| Token top-up | planned | Enable APIKEY as a payment path in the console, with the 20% burn live from the first transaction. |
| Staking launch | planned | Deploy the mined address once a reward period is funded by vote. |
| Independent audit | open | Third-party review of the five deployed contracts before token payments carry material volume. |
The ordering is deliberate. Governance before token payments, funding before staking, audit before volume. Each step is a prerequisite for the next rather than a parallel workstream, because the failure mode of shipping all of it at once is that nobody notices which piece is broken.
This document describes software that exists and numbers that were measured. It does not forecast a price, promise a return, or claim that the token will appreciate. Anyone reading it as an investment case is reading it wrong.