Overview & result
Prism is an autonomous liquidity management agent for decentralized exchanges. Rather than delegating order execution directly to a language model, Prism uses Jev as a high-speed market regime classifier that identifies toxic order flow, directional pressure, and volatility shifts, handing regime states off to backtested mathematical strategies.
From the original post
Autonomous LP trading agent - auto-rebalancing with backtested strategies - Currently support Meteora DLMM. Doesn't directly have Jev place orders; judges states like toxic flow, market pressure, mean reversion.
How Jev fits in the loop
- Ingests live pool order books, trade sizes, and liquidity bin distributions.
- Queries Jev for fast regime classification (e.g., toxic directional flow vs mean-reverting chop).
- Dispatches deterministic rebalancing instructions to DLMM pools based on the classified regime.
How to reproduce
- Clone https://github.com/irfndi/prism-liquidity-agent and install packages.
- Set up testnet RPC endpoint and TypeSafe API credentials.
- Run simulation on Meteora DLMM pools to observe Jev classifying toxic flow and directing liquidity rebalancing.
Why this build matters
Bridges AI decision classification with deterministic DeFi market making, avoiding hallucinated transaction parameters while reacting to microstructure changes.
Limitations
- DeFi trading carries inherent smart contract and impermanent loss risks.
- Currently tailored to Meteora DLMM pool architectures.