Racksx Phase 1 of 3 training RugAI
— 0 rep 50
RUGAI v7 · LIVE Tokens labelled 17,000 Accuracy 24 h 90.7 % AUC-PR 24 h 0.965 Validators 212 Pool left 175.0M Leverage waitlist 1,284 demo data
The memecoin leverage broker

Racksx

Real loans · real tokens · your wallet
Scroll
The memecoin leverage broker · Solana

Leverage on pump.fun memecoins. Real loans, real tokens, your wallet.

Lenders fund the loan. The token you buy is the collateral, locked on-chain until you repay. RugAI decides which tokens can be lent against, how much, and sells before the crowd when a rug starts.

Join the leverage waitlist Start validating

No custody. PnL comes from the real market on PumpSwap and Jupiter, never from a house against you. Leverage goes to mainnet only after the model passes and the program has 90 % test coverage.

Phase 1 is live

RugAI. The broker's brain, built in public.

Before we lend a single dollar, our AI has to earn it. Paste any pump.fun link and see what it thinks. Help train it. Get paid.

16,800
tokens analysed
90.7 %
accuracy, last 24 h
204
active validators
1,240
on the leverage waitlist

Free, no wallet. Not financial advice. The model is wrong 9.3 % of the time on the last 24 hours, and a low probability is not a guarantee.

Three products, one order

The shield first. Then the terminal. Then the loan.

The broker cannot lend against a token until the model that scores it is trained and evaluated. That is a hard gate, so the AI ships first and the community trains it in the open.

1 · NOWlive

RugAI

The probability that a memecoin is a rug pull, explained signal by signal, measured in public every day. The community validates the outcomes and earns tokens for it.

  • Risk at 1 min, 5 min and latent
  • Real outcome at 1 h, reviews at 24 h and 7 d
  • Daily retrain, champion vs challenger
Holding unlocks unlimited checks, latent-risk alerts, the Telegram bot
2 · NEXTphase 2

Spot terminal

Buy and sell any pump.fun token from Racksx with limit and stop orders and an exchange feel. The wallet stays yours. First revenue line: 0.75 % per trade.

  • Bonding curve direct, graduates via PumpSwap or Jupiter
  • Card on-ramp to your own wallet
  • No audit needed: nothing is custodied
Holding unlocks a reduced fee, 0.5 % instead of 0.75 %
3 · THE COREphase 3

Leverage with a real loan

Lenders put USDC in a pool and earn interest plus a rug premium. You buy the real token with your money plus the loan. It stays locked as collateral. 1.5x to 5x by tier.

  • RugAI sets tier, loan cap and leverage per token
  • Auto-close when the creator's cluster sells
  • Insurance fund covers bad debt in the same transaction
Holding unlocks the private beta whitelist and lender priority
more checks→more labels→a better shield→safer loans→more traders and lenders→more checks
Why a shield at all

The classic rug is gone. Hidden concentration replaced it.

pump.fun burns liquidity at migration, so nobody pulls it any more. The creator, linked wallets, block-0 bundles and snipers buy cheap and sell on retail in the first minutes. Scanners check what the protocol already fixed.

98.6 %of 7 million+ tokens launched on pump.fun in 2024–2025 ended with no liquiditySolidus Labs, 2025
36.5 %of supply in coordinated accounts that look independent, across 41,000 graduated launchesMELT, Georgia Tech, 2026
0.96AUC-PR of RugAI on temporal validation. The best published academic model reports 0.80Catching the Rug, 2026
PLANTYHIGH
100 %

Risk at 5 minutes. 85 % already at 1 minute: the deployer had 32 previous tokens, all dead.

Real outcome: rug by sniper dumpPredicted before any trading. Would never be listed with leverage.
COCKROACHMEDIUM
37 %

154 buyers in 5 minutes, price ×17, migrated to PumpSwap in 12 seconds. A traditional scanner would have shouted rug because of the non-standard supply.

Real outcome: legitStill above 40 %, so spot only until the probability falls.
How RugAI learns

The community labels what the AI predicted. The AI retrains daily.

Paste a linkor watch the live feed
Read the chain123 signals, clustered
Risk probability1 min · 5 min · latent
Real outcomeauto-labelled at 1 h
Community votesRug · Legit · Unsure
Consensus ≥ 70 %label closed, matching votes paid
Daily retrainchallenger vs champion

A challenger model retrains on the new labels every day and replaces the champion only if it does no worse on the last 24 hours of live tokens. Nobody has to tell the system what a rug is.

Roadmap

RugAI first, then spot, then leverage on devnet, then mainnet in stages.

Each phase opens the next only when its gate is met. The two hardest gates come from the broker spec: no lending against a real token until the model passes, and no leverage on mainnet until the program has 90 % test coverage.

1 · RUGAIPublic modelVotes, reputation, leaderboard, token launchGate: 200 validators, token launched
2 · SPOTMainnet terminalOrders, portfolio, 0.75 % fee, holder discountGate: real swaps, fees collected
3 · MARGINOn devnetLoans, pool, insurance fund, oracle, shieldGate: 90 % test coverage, rug sim closed
4 · BETAPrivate leverageWhitelist, tiers A and B, 2x, 25k USDC poolGate: 4 weeks without bad debt
5 · PUBLICStaged leverageA–C at 3x, then A–D at 5x, 500k poolNo absolute caps after an external audit