sambhal-labs
LiveFirst product

Mayavi

Forked-mainnet simulation engine for token DD and protocol risk.

Replay token launches, depeg cascades, and liquidation events against real on-chain state — bit-exact (delta == 0 against Uniswap V3 Quoter, validated on the EIGEN airdrop). On top of that engine, we train RL agents: a v2 PPO Aave borrower outperforms a passive baseline by +18 bps. Self-serve through app.mayavi.sambhal-labs.com.

Visit Mayavi Open the dashboard

Three pillars

Bit-exact validation

delta == 0 on the 7-swap Uniswap V3 Quoter suite. EIGEN incident replay matches on-chain truth at the airdrop block. Methodology is public at docs/validation.md.

RL-trainable agents

Train PPO against the same forked-mainnet env you use for DD. Aave borrower (Modal A10G v1 +18 bps; local v2 +3.35 bps; v3 200K-step saturation finding), vesting recipient, and Aave liquidator all ship with published evals — including the negative results.

Self-serve, multi-chain, open

Submit through the dashboard, get a report in minutes. 6 EVM chains (mainnet, Arbitrum, Base, Optimism, Polygon, BNB) live; Solana via Surfpool parked. Methodology is open and auditable. Single API key for v0.

Who it's for

Five buyer types, ordered by where Mayavi fits today.

01
Protocol risk team
today

“We're considering raising the Aave LT for ETH from 80% to 82%. What's the cascade risk under a 30% drawdown?”

aave_borrower + depeg_cascade + cross_protocol_shock

02
VC token-DD analyst
today

“Token unlock in 14 days. What's the realistic sell pressure given current pool depth?”

vesting_cliff + launch_replay

03
Market maker
today

“ENA unlocks $400M in 2 weeks. Where do we want to sit?”

launch_replay parameterized + custom timing

04
DAO treasury manager
today

“Reserves are 60% in our own token. Stress-test under a 50% drawdown + 30-day liquidity drought.”

multi_cohort + depeg_cascade + custom treasury agent

05
DeFi RL researcher
today

“I want to train a liquidator policy against a forked-mainnet env, gymnasium-compatible, with reproducible bit-exact rollouts.”

gym_env + Ray RLlib

RL agent matrix

What's shipped, what's in flight, what's on the roadmap. Same engine, different agent classes.

Aave borrower under shock
shipped

PPO trained on Modal A10G (v1 +18 bps) and local GTX 1650 (v2 +3.35 bps); v3 (200K steps) surfaces an honest saturation finding — full eval JSONs published.

Vesting recipient
shipped

Heuristic urgency_beta sell schedule plus a trained PPO eval against twap / dump_all baselines; saturation regime documented.

Liquidator
shipped

PPO eval against the scripted close-factor heuristic — PPO captures 63% of the heuristic's realised gas-adjusted return. Negative result, published as is.

Multi-borrower cascade coordinator
roadmap

Multi-agent RL — first such system trained on real EVM

AMM LP (concentrated liquidity range)
roadmap

Optimal tick range under volatility regime X

Treasury manager
roadmap

Backtested rebalance trigger against on-chain history

Get involved

We're onboarding our first design partners. Free for the next 90 days, weekly feedback only — no money, no commitment beyond a monthly call.

Become a design partnerRead the docs