Live mode pulls real pairs from DEX Screener, exactly as they trade. Switch to the troop replay to watch ENTRY catch a pair, FILTER recheck every wallet, RISK ask for an exit rule and STOP and TAKE agree on one number.
Troop screenerSIM
ETH · WETH pairsSource: DEX Screener APIT+00:00Loading…
Pair
Price
5m
1h
Vol 1h
Liquidity
Chart
Troop
Verdict
Watched-wallet tape
Caught
0
BURST
0
Went 2x
0
To RISK
0
Vetoed
0
Live mode shows DEX Screener data as it is, not picks or signals. In the troop replay the pairs, prices, wallets and verdicts are simulated.
Each monkey got a job, not a prompt.
Pick a monkey to see its job at the desk.
entry.desk0
sim view
109
Genomes competed
8,230
Tokens scouted
2,023
Simulated fills
+1,071%
On the sim book
The coin
MONKEY
One coin, seven monkeys, one desk. Drag it, flick it, flip it. This one is a toy: it spins, it lands, it tells you which side is up and nothing else.
One objective. Seven agents. One surviving strategy.
7 MONKEYS is an autonomous strategy-evolution desk built on GPT-6 Astra. It takes a single objective, splits it across seven specialist agents, and lets a population of trading genomes compete under identical conditions until one survives fees, drawdown and data it has never seen.
Every genome starts with the same $1,000 in virtual capital. Nothing is hand-tuned between rounds. The operator sets the objective and takes their hands off the keyboard.
02 / Input
The troop receives one sentence, not a prompt library.
A run needs an objective and a tape. Everything else is evolved. The desk reads public on-chain events directly; it never asks for a browser wallet, a signature or an approval.
Objective
One sentence. The only instruction the troop ever gets — e.g. “evolve a trading strategy that survives fees, drawdown and unseen data.”
Capital
$1,000 virtual per genome. Identical for every competitor, every round.
Genes
Seven trading genes per genome: entry · risk · stop · take · size · hold · filter. Each is owned by one monkey.
Tape
Public on-chain activity: 8,230 tokens and 250 top wallets on the last run. Read block by block, not scraped from a leaderboard.
Holdout
A slice of history the troop never sees while it competes. The winner is locked before the holdout is opened.
03 / The receipt
Every trade keeps its reason attached.
A result never shows a number alone. The observed market state, the genome that traded it, the fee assumptions, the simulated fills and the drawdown stay visible together. Four checks stand between a candidate and the book:
01
First lookENTRY
The scout screens the tape and hands candidates down the chain. A wallet buying something is an event, not a thesis.
02
Exit ruleRISK
No entry without an exit. Every veto comes back with a reason written into the ledger — 11 of 412 handoffs on the last run.
03
One numberSTOP · TAKE
The brake and the closer argue over simulated fills until they agree on a single exit level before the token is named.
04
Full runHOLDOUT
The locked winner is replayed on data it has never seen. Returns are reported after fees and drawdown, not before.
04 / Method
Selection, mutation, extinction. Then the test.
Genomes compete on the same tape with the same capital. Survivors reproduce; a child inherits its parent and mutates one to three genes. Failed genomes are archived, never deleted, so every lineage can be traced back to its first ancestor.
Fitness is measured after costs. A strategy that only works with a perfect entry or one lucky trade is sent back for another round of calculations.
Gross return+ P&L
−
Feesevery fill
−
Drawdownworst path
=
Fitnessafter costs
BURST rule
When four wallets scoring 80+ buy the same token inside 30 minutes, the desk fires a BURST. On the last run 44% of those signals went 2×. The rest were handed to RISK and never reached the book.
05 / The troop
Each monkey got a job. Not a prompt.
Seven specialists, constant disagreement, one reason behind every trade. No agent can both propose and approve the same action.
#
Agent
Role
What it owns
01
ENTRY
The scout
Screens tokens and top wallets on chain; finds the signal before the troop makes a move.
02
RISK
The skeptic
Vetoes any entry without an exit rule and writes the reason into the ledger.
03
STOP
The brake
Sets the level where a trade is wrong. Argues with TAKE until they agree.
04
TAKE
The closer
Sets the level where a trade is done. Agrees on one number before ENTRY names a token.
05
SIZE
The allocator
Routes capital between surviving genomes. Nothing gets real size before the holdout.
06
HOLD
The bookkeeper
Keeps the ledger: every fill, every veto, every mutation, every archived failure.
07
FILTER
The gatekeeper
Rechecks every wallet before the troop is allowed to follow it. Runs the tail sessions.
06 / Safety boundary
Read-only by design. Simulated by default.
The desk observes public events and runs its competition in a simulation. It needs no wallet connection and never requests signatures, approvals or private keys.
No wallet
Nothing on this site or the desk asks you to connect, sign or approve anything. If something does, it is not ours.
Not a quote
Simulated fills and returns are estimates under stated assumptions — not executable quotes, not a track record, not financial advice.
Not a promise
Figures are supplied by the creator from an independent experiment. Past simulated performance does not predict future results.
I set the objective. The monkeys divided the work, killed each other's bad ideas and came back with one strategy and a reason for every trade.