Put every Solana memecoin on trial before it puts your wallet in the ER.
coinjury pulls the memecoins trending right now, runs each one through real
on-chain rug checks, and hands down a verdict — WATCH / COINFLIP / AVOID.
Then it can run a transparent $20 experiment: let an AI make the calls and log
exactly how it does, on a public scoreboard, over time.
No API keys required. Every data source is keyless, and if one is down the tool degrades gracefully instead of breaking.
Not financial advice. coinjury reads public data, never touches your keys, and never tells you to buy anything. Memecoins are gambling — most go to zero. See DISCLAIMER.md.
Can an AI flip $20 in Solana memecoins without getting rugged?
Most memecoin "AI bots" are black boxes that promise gains. coinjury does the
opposite: it starts with $20 of paper money, lets the judge open small positions in
the least-bad names it can find, marks them to live prices, and commits the whole
thing to SCOREBOARD.md — wins, losses, and all — on a daily
schedule. The point isn't to get rich. It's to find out, in the open, whether
safety-first + AI beats a coinflip. (Spoiler: memecoins are gambling. This is
tuition, not a strategy.)
Bonk Bonk
verdict WATCH risk 18/100 [###-----------------]
liq $333k vol24h $762k fdv $304.4M age 32429h
No glaring killers, but FDV > 100x liquidity (tiny float, easy dump). 24h +5%. Still risk-only money.
on-chain checked: rugcheck + rpc + jupiter | rugcheck 7/100
RUGME Rug Me Daddy
verdict AVOID risk 96/100 [###################-]
liq $2k vol24h $90k fdv $400k age 20h
Hard pass. mint authority active (dev can infinite-mint); liquidity under $2k. 24h +120%.
Critical gates floor the score at AVOID no matter how good everything else
looks. Any one of these and the token fails:
- mint authority still active — dev can infinite-mint and dilute you to zero
- freeze authority still active — dev can freeze your tokens so you can't sell
- no sell route — you can buy but can't sell (honeypot)
- transfer tax > 10% — a soft honeypot
- RugCheck marks it already rugged
Weighted penalties then sum on top: LP not burned/locked, ruggable liquidity, holder/market froth (volume ≫ liquidity, tiny float), violent price swings, and brand-new pools.
Signals come from public, keyless sources and cross-check each other:
| source | gives |
|---|---|
| Solana JSON-RPC | mint / freeze authority — ground truth, read straight from chain |
| RugCheck | LP-locked %, transfer tax, "rugged" flag, holder/insider risks, composite |
| Jupiter | a real sell-route quote — can you actually get out? |
| DexScreener | liquidity, FDV, volume, age (discovery + market data) |
When two sources disagree (e.g. RugCheck says renounced but the chain says active), coinjury trusts the chain and flags the mismatch. Safest-first ordering is about survival odds, not upside — and the tool never tells you to buy.
git clone https://github.com/Overdusts/coinjury
cd coinjury
python coinjury.py scan --demo # try it offline on bundled data, no setupNo dependencies to install — coinjury runs on the Python standard library.
python coinjury.py scan # trial the current trending memecoins (live)
python coinjury.py scan --demo # bundled sample data, offline, no calls
python coinjury.py judge <MINT> # full rap sheet on one token
python coinjury.py challenge # run / refresh the $20 paper experiment
# options
--limit N how many to screen (default 12)
--min-liq USD ignore pools below this liquidity (default 10000)
--no-enrich skip on-chain checks (faster, market-data heuristic only)
--ai richer verdicts via an LLM if ANTHROPIC_API_KEY is set
--json machine-readable output
--versionBy default the verdict text is written by a local rule-based judge — zero keys,
fully offline. Set ANTHROPIC_API_KEY and pass --ai to have a model write a
blunter take instead. It falls back to the local judge on any error, so the tool
never breaks without it.
python -m unittest discover -s tests -vOffline unit tests cover the scoring gates, penalties, verdict bands, graceful degradation, and the full demo pipeline.
Be honest with yourself about the limits. coinjury cannot detect off-chain team intent, a slow rug (a dev who dumps next week), a coordinated social pump, a future CEX listing, or a brand-new scam pattern its rules don't encode yet. It measures on-chain and market survival odds at one moment — not the future, and not upside.
coinjury collects nothing about you: no analytics, no telemetry, no account. It
runs entirely on your machine. In live mode it sends token mint addresses
(public info) to the public APIs above to fetch data; --demo makes no network
calls at all. With --ai, token metrics are sent to Anthropic. See
SECURITY.md.
This is a research and entertainment tool. It logs how an AI performs; it does not recommend anything to you. Memecoins are among the riskiest things you can touch — the overwhelming majority go to zero. Never put in money you can't afford to lose. Full terms in DISCLAIMER.md.
MIT © Overdusts