how d1 decides
d1 is a decision model by Liquid AI. It does not write text. It reads a coin once and returns probabilities for a fixed set of answers in a single pass. That is why a whole read takes about 378ms, network included.
new coins
A GMGN feed of new pump.fun coins, polled every few seconds. Only coins that no tracked KOL has bought yet.
- age 20s to 600s
- market cap $5,000 to $30,000
- paired with SOL, launched on pump.fun
code filters
Numbers are cheaper to check in code than in a model. A coin stops here if any of these fail.
- rug ratio at most 0.3
- bundled supply at most 30%
- insiders at most 20%
- top 10 wallets at most 50%
- creator holds at most 15%
- dev team at most 20%
- at least 8 holders
- creator made at most 50 coins before
one pass through d1
Every coin that passes goes to d1 with a compact state: name, ticker, socials, cap, holders, buys and sells, creator history and holder quality. d1 answers all five questions at once.
A fast trader buys new coins only before notable traders arrive and sells into their buying. What should the trader do with this coin right now? enter, watch or skip
How likely is it that well known memecoin traders (KOLs, smart money) will buy this new pump.fun coin within the next few minutes?
How many notable traders are likely to pile into this coin soon? none, one or several
Is this a low effort or copycat coin (generic name, duplicated image or socials, serial creator)?
Does this coin look like a rug or bundle setup (creator or insiders hold a lot, bundled supply, serial creator whose coins never graduate)?
enter
d1 buys only when every answer clears its line.
- buy now at least 0.14
- kol pull at least 0.6 of 4
- rug risk at most 0.3
- low effort at most 0.75
0.02 SOL a trade, at most 3 open at once, new entries stop after 0.15 SOL lost in a day. Right now it trades on paper.
see every tradeexit
The first rule that fires closes the whole position.
- a tracked KOL buys the coin: sell 5s later, into their demand
- take profit at 1.5x
- stop loss at minus 25%
- no KOL after 180s: sell
why it sells into KOLs instead of copying them
Before building d1 we pulled four days of trades from the top 100 KOL wallets on GMGN: 48,000 trades across 4,764 coins, with price candles for every coin.
So d1 tries to be early, not to follow. Its job is to guess which brand new coins will pull notable buyers, and to leave when they arrive.
the model
d1 by Liquid AI: a 3B decision model built on LFM2.5. It scores options by reading the answer slot once, so it costs about $0.00002 a coin and never writes a word. Your own d1 bot can add up to five plain English yes or no rules, answered in the same pass.