Skip to content

AI Advice

/advice

Bundles every metric in the Metrics Catalog for each of your open positions, sends them to an LLM along with global market context (ETH/BTC indicators, Fear & Greed), and returns per-position recommendations.

What the advisor sees

For every open position, the LLM receives:

  • Position core — ID, pair, chain, P/L vs entry, P/L vs HODL, liquidity value, uncollected fees, gas cost estimate, age, in-range flag, pair RSI(14).
  • Pool snapshot — TVL, 24h volume, 24h fees, fee tier, tx count, AI risk classification with rationale, recent open/close activity.
  • Similar pools — DefiLlama matches with the same pair / fee tier on other protocols, used as a yield benchmark.
  • Range analysis — distance to bounds, annualized pair volatility, random-walk ETA to each bound, daily fee run-rate, days to breakeven, and a will_likely_exit_before_breakeven flag.
  • Market context — ETH and BTC 24h indicators, Fear & Greed Index.

The full schema is documented in the Metrics Catalog.

Acting on advice

Every recommendation comes with one or more inline buttons to prepare a transaction for the suggested action — collect fees, mint a new position with adjusted bounds, or close the position.

Tapping a button asks the bot to build an unsigned transaction. The bot returns the calldata and asks you to sign and broadcast from your own wallet. YummyBait never holds keys and never broadcasts on your behalf.

Caveats

  • Not financial advice. The advisor is a tool, not a recommendation. Verify every action before signing.
  • Latency. A single /advice call can take 10–30 seconds depending on how many positions you hold and current LLM load.
  • Stale on-chain state. Indexer lag is typically <1 block, but if you've just made a transaction, give it a moment to settle before asking for advice.