A Layer-1 network where an AI answer is not something you are asked to trust, but something anyone can check.
When a machine-learning model produces an answer, the person receiving it has no way to confirm what actually happened. Was that model really used? Was the output altered afterwards? Was the same question answered differently for someone else? Today the only available answer is: trust the operator.
For casual use that is acceptable. For anything consequential — a score, a verdict, a classification that decides an outcome — it is not. And it becomes untenable the moment a smart contract, rather than a human, is meant to act on the answer.
Exavolt Chain executes each inference job twice, on two independent staked node operators. Both must produce byte-identical output. Each output is hashed with keccak256, and the two hashes are compared on-chain.
A result is accepted only when the hashes match. The comparison is a pure function of stored state, so anyone can trigger and verify it permissionlessly — the operator is not required to be online, cooperative, or honest for the check to work.
If the two outputs disagree, no result is delivered. The job escalates to dispute, and the funds involved retain a permissionless exit path.
Byte-identical output has been reproduced across different physical machines, operating systems and CPU models — same model, same input, same bytes. Redundancy is not overhead here; it is the mechanism.
Matching output proves a result was produced by the specified model and input, and was not altered afterwards. It does not prove the answer is correct. Models can be wrong, and models can be misled.
What Exavolt offers is not an AI that cannot err. It is an error that cannot be hidden.
| Parameter | Value |
|---|---|
| Network name | Exavolt |
| Chain ID (EVM) | 188188 · 0x2DF1C |
| Chain ID (Cosmos) | exavolt_188188-1 |
| Native token | EXVT — 18 decimals |
| Base denom | aexvt |
| Block time | ~3 seconds, deterministic finality |
| Consensus | CometBFT (Tendermint BFT) |
| Execution | EVM-compatible — Solidity, EVM target paris |
| Total supply | 1,000,001,000 EXVT — fixed |
| Inflation | Permanently disabled at genesis |
| RPC | https://rpc.exavolt.io |
| Explorer | https://explorer.exavolt.io |
Three layers, each doing one job:
The chain does not run the model. The model runs off-chain on operator hardware; the chain records what was asked, what was returned, and whether independent operators agreed. That separation is what keeps verification cheap and the ledger small.
EXVT is the native gas token. It pays for computation and compensates node operators for verified inference work.
The project does not sell EXVT.
The project does not operate or maintain a market for EXVT.
The project does not set, quote, or project any price.
EXVT is obtained by performing work on the network. There has never been a presale, a private sale, or an allocation to any investor.
This document describes how the network functions. It is not an offer, a solicitation, or investment material.
| Allocation | Amount | % | Purpose |
|---|---|---|---|
| Ecosystem | 300,000,000 EXVT | 30% | Grants, node staking, ecosystem support |
| Rewards pool | 250,000,000 EXVT | 25% | Node operator compensation |
| Development | 200,000,000 EXVT | 20% | Building and maintaining the network |
| Contributors | 150,000,000 EXVT | 15% | Compensation for work performed |
| Reserve | 100,000,000 EXVT | 10% | Contingency |
| Validator (staked) | 1,000 EXVT | <0.01% | Consensus |
| Total | 1,000,001,000 EXVT | 100% | Fixed — no further minting is possible |
Allocations are purpose-bound and held in a 2-of-3 multisig. Balances and every movement are publicly readable on the explorer. Allocation labels are a stated policy, not yet a contract-level constraint — see Current status.
| Property | Implementation |
|---|---|
| Redundancy | Two independent staked operators per job |
| Agreement | keccak256 of both outputs compared on-chain |
| Verification | Permissionless — a pure function of stored state |
| Disagreement | Escalates to dispute; no result is delivered |
| Fund safety | Every non-final state has a permissionless exit — funds cannot be held hostage by any party, including the project |
| Privileged roles | 2-of-3 multisig across three independent devices, including a hardware wallet |
| Contract authenticity | On-chain registry — isOfficialContract(address) |
The last row of that table is the one that matters most in practice: a contract that holds user funds must never have a state from which those funds cannot be recovered without someone's cooperation. That property is enforced in code, not by policy.
The registry is the authoritative source for which contracts are official. Verify addresses against it — not against social media, third-party lists, or this document.
getAllContracts() returns every address ever registered, including deactivated ones. It is a history, not a list of currently official contracts. The only authoritative check is isOfficialContract(address), queried per address.
Licensing: licences differ per contract. Each deployed contract carries its own SPDX identifier in its verified source on the explorer — check there rather than assuming a single project-wide licence.
Stated plainly, because a technical document that only lists strengths is not a technical document.
Every claim in this document that concerns on-chain state can be checked directly: query the RPC, read the verified source on the explorer, call the registry. If a statement here cannot be verified that way, treat it as an intention rather than a fact.