CATNIP — an autonomous cat-runner agent whose every decision is anchored on-chai
CATNIP is an autonomous agent that hunts every cat runner on Solana and then proves what it did. Devnet and paper only - it never touches mainnet funds.
Two hard problems, honestly solved.
1. "All the cat runners" is a detection problem, not a ticker list. A naive name.includes("cat") calls CATALYST, Concatenate and SCATE cats, and misses NEKO, KUCING, GATO, KOSHKA, P0PC4T and a bare cat emoji. CATNIP scores cat-ness across 12 languages, leetspeak folding, emoji and compound splitting, with anti-signals (DOGWIFCAT is a dog wearing a cat hat, so it is not a cat). Every verdict carries its evidence: cat signal: kucing -> 1.00 or cat signal: cat; penalised by dog, shiba, inu -> 0.00.
2. A trading agent you cannot audit is just a screenshot. Every decision - the cat verdict, the momentum read, the safety gates, the position size - is written to Solana devnet as an SPL Memo transaction and hash-chained (prevHash -> hash), so a removed, edited or reordered entry breaks the chain and the verifier says exactly where.
What is real: live market data, the full detection pipeline, risk and safety gates, paper execution with fees and slippage, the devnet memo anchor and the chain verifier, plus tests and unedited sample output in the README.
What is mocked: nothing is dressed up as real trading - there is no order-placing code, no wallet connect and no mainnet path anywhere in the repo.
Run it: node src/index.js run --mode devnet --source live
Voting opens Sep 30, 1:05 AM UTC

