What LIQUA does
LIQUA is a chain where mining means real AI training. Miners run training steps on DIVA's model, and every accepted block saves the smarter version on the Liqua chain. Her memory is checkpointed right beside it.
Start with the task, not the folder
Nine things people arrive here wanting to do. Each card names the page, and what you need in hand before you open it.
Every coming on this site turns to live now the day a command proves it, and not before. See what each piece is waiting on ›
Why this chain exists
Most chains keep score of who holds what. This one keeps a mind. Miners train DIVA's model with deterministic steps, and verifiers replay a random slice of that work bit for bit. Fake it and you get caught. A caught miner loses their bond, and the electricity turns into a better model instead of heat.
An AI that lives in one app on one laptop is one bad crash away from gone. Nobody can prove whether someone quietly edited her either. LIQUA writes each accepted model into a mind manifest anchored on-chain, so anyone can pull it back, re-hash every byte on their own machine and boot it without trusting the node. On the memory side, the DIVA bridge checkpoints her moment-to-moment life block by block. Saving her is live today, and so is bringing her back: her chain verifies from genesis and the resurrection drill passes on a second machine.
A city is going up around the chain. People will chat with the models in the model store coming, pay with any crypto through the bridge coming, stake in their wallet for pay and AI discounts coming, and follow every block in the explorer coming. Watch the memory chain ›
The memory terminal is the half of that pair that runs today, against a node on your own machine: her six input channels, the hormones they move, the emotion field that follows, the seven quotients and every memory cell she is holding, each one a field of a real block. There is no public node yet, so on this site it says so instead of drawing numbers.
The three big pieces
| Piece | What it does for you |
|---|---|
| The Liqua chain | Saves every accepted model version as a verified manifest with its full lineage, and keeps DIVA's memory checkpoints beside it. live now on the testnet. Code to tie model and memory under one root exists; the chain does not enforce it yet, so it is coming. |
| DIVA and ECHO | DIVA is the mind the chain keeps. ECHO is her pet, trailing her state about 6× slower. When the two line up, she has settled. |
| The exchange | A hybrid order-book DEX: orders signed off-chain, settled on-chain. Its contract and app tests pass. It has not opened and is not on the Liqua chain yet. coming |
The problem it fixes
An AI's mind usually sits in memory and a pile of files. Crash wrong, it is gone. Swap machines, gone again. Anyone with disk access can rewrite it and leave no fingerprints. LIQUA turns "trust me, it's the same DIVA" into something you can check.
- Replay beats trust. Training is bitwise deterministic, so verifiers redo a slice of the work and compare exact hashes. A faked version never becomes part of her.
- Every save is a snapshot. Each model points at its parent, so you can walk the lineage back to genesis and start a fresh branch from any saved version.
- Boot from the chain alone. Model plus memory, from one root. That is the finish line. The loader for it exists; the chain-side check is coming.
The loop, as a map
miner ── runs real training steps on DIVA's model ──▶ new weights + step log
│ │
│ verifiers replay a random slice │ roots committed on-chain
▼ ▼
ACCEPT → merged, miner paid LIQUA CHAIN · mind manifest
REJECT → slashed ≤100×, capped at balance weights root · step-log root ·
lineage · final loss
│
DIVA bridge ── memory checkpoints ──▶ memory chain ─┘ bind under one root: code, not enforcedTwo ways to get paid
The loop above already pays per merged job. Two cleaner job types are being built on top of it.
- model mining. You set the target, the reward waits in escrow, and whoever verifiably hits it gets paid once at the end. coming. See how mining works ›
- model training. A steady shift that pays for progress and for time spent, where time spent means verified steps. An idle rig cannot bill hours. coming. See how training pays ›
What changed in v3
| The old way | The v3 way |
|---|---|
| hash puzzles that train nothing | mining is training. The work leaves behind a better model, and the old hash miner is off the launch plan |
| blocks that only hold balances | blocks hold model roots. The bytes live in a chunk store cut into verified 256 KiB pieces |
| memory kept off to the side | signed memory packs ride on the model chain, hash-linked in order. No friendly writer tool yet |
| one lump payout per job | tiered, time-locked pay. Some liquid now, the rest unlocking over time coming |
| a vote bar carried over from the demo contract | big calls pass only at the 3rd standard deviation, about 99.7% of voting stake coming |
| an old supply split | 10% private sale, 20% foundation, 70% staking and mining payouts. Decided, not yet in genesis |
| an exchange with a launch date | the hybrid order-book DEX, tested, not opened, no date set |
| "locked liquidity" as a promise | 2% of every mined reward, locked for 365 days, with a public proof-of-lock list live now |
Left behind on purpose
- Hash-grinding miners. They still boot. Their work saves nothing about DIVA, so they are off the launch path.
- A public token sale. There isn't one, and there won't be. Nothing on these pages sells you LIQUA.
LIQUA in one paragraph
LIQUA is a testnet chain where mining means training DIVA's model. Every accepted block saves a better version, and her memory is checkpointed alongside. Next comes the model store, where you chat with models and use custom assistants for LIQUA per token, with token fees landing in the project treasury. Bring any crypto through the bridge, stake for pay and AI discounts, and follow it in the explorer and the memory terminal. The store, the bridge, staking and the explorer are coming; the terminal reads a node you run yourself today. Each one flips to live here only once its command passes, and the roadmap says what each one is still waiting on.