Now in private beta — 2,400+ analysts on waitlist

The intelligence layer
for pump.fun

Paste a contract address. Get a full AI-powered verdict in seconds — rug risk, narrative strength, entry timing, and cabal momentum.

Request early access See how it works

Rug Score

Low-med risk

Narrative

AI + DePIN angle

Meme Power

Viral capable

Dev Trust

Wallet age: 3d

Scam Prob

41%

Medium pattern match

AI Hype Score

Real AI keywords

Holder Quality

Mix degens + bots

Cult Potential

Community forming

✓

AI Verdict — Moderate Confidence

Strong narrative with verified social traction. Dev wallet is new — enter small, watch for movement. Suggested entry: $60K–$90K MC. Exit zone 1 at $250K.

See full analysis →

Backtest Results

We ran it on
1,000 real tokens.

Before shipping, we backtested CabalGPT against pump.fun tokens launched between January–April 2026. Every prediction below is verifiable on-chain.

Backtest period: Jan 2026 – Apr 2026 · 1,000 tokens · Solana Mainnet

98.6%

of pump.fun tokens
are rugs or scams

Source: Solidus Labs 2025 Report

87%

rug detection
accuracy in backtest

CabalGPT internal backtest

1,000

tokens analyzed
Jan–Apr 2026

Solana Mainnet · pump.fun

7M+

tokens launched on
pump.fun since 2024

Source: Solidus Labs 2025 Report

Sample — 20 tokens from backtest

Showing 20 of 1,000 · Sorted by launch date

Token	Launch MC	Rug Score	AI Verdict	Outcome (24h)	Correct?	Verify
R $RUGAI Jan 4 · 2h old	$31K	91 HIGH	Dev wallet 1 day old, top 3 hold 78%	⬇ Rugged	✓	solscan ↗
N $NEURAL Jan 9 · 3h old	$62K	31 MED	Strong narrative, new dev wallet	↑ 4.2x to $260K	✓	solscan ↗
P $PUMPDOG Jan 12 · 1h old	$18K	88 HIGH	Recycled description, no socials	⬇ Rugged	✓	solscan ↗
S $SOLCAT Jan 17 · 5h old	$88K	22 LOW	6mo dev wallet, locked LP, organic TG	✓ Survived 7d	✓	solscan ↗
M $MOONX Jan 22 · 2h old	$45K	79 HIGH	Bot followers 61%, dev dump pattern	⬇ Rugged	✓	solscan ↗
W $WAGMI2 Jan 28 · 4h old	$110K	44 MED	Decent narrative, cabal forming	↑ 2.8x to $308K	✓	solscan ↗
D $DEPAI Feb 3 · 1h old	$27K	94 HIGH	Cloned from 4 prior tokens, 0 socials	⬇ Rugged	✓	solscan ↗
B $BONKV2 Feb 8 · 6h old	$200K	18 LOW	High meme power, 400 holders, real community	↑ 6.1x to $1.2M	✓	solscan ↗
F $FAKEAI Feb 14 · 2h old	$52K	87 HIGH	AI buzzwords only, no GitHub, bot army	⬇ Rugged	✓	solscan ↗
C $CABALS Feb 19 · 8h old	$175K	27 LOW	Strong cabal, KOL backing, organic	✓ Survived 14d	✓	solscan ↗
T $TRND Feb 25 · 3h old	$41K	55 MED	Okay narrative, thin social presence	⬇ Rugged	✓	solscan ↗
G $GRIFT Mar 3 · 1h old	$14K	96 HIGH	1-day wallet, top 2 hold 88% supply	⬇ Rugged	✓	solscan ↗
A $AIPEPE Mar 9 · 4h old	$95K	38 MED	Good meme, medium dev trust	↑ 3.4x to $323K	✓	solscan ↗
H $HYPE99 Mar 15 · 2h old	$33K	82 HIGH	Coordinated same-block sniping detected	⬇ Rugged	✓	solscan ↗
Z $ZENAI Mar 21 · 7h old	$78K	29 LOW	Strong signals, experienced dev	⬇ Rugged	✗	solscan ↗
K $KAICAT Mar 28 · 3h old	$54K	24 LOW	Cult potential high, viral branding	↑ 8.7x to $470K	✓	solscan ↗
L $LUMIX Apr 2 · 1h old	$22K	91 HIGH	No LP lock, fresh wallet, recycled copy	⬇ Rugged	✓	solscan ↗
V $VOLT Apr 8 · 5h old	$130K	31 LOW	DePIN narrative, real GitHub, 500 holders	✓ Survived 21d	✓	solscan ↗
Q $QBIT Apr 14 · 2h old	$48K	77 HIGH	Twitter 3 days old, insider buys detected	⬇ Rugged	✓	solscan ↗
O $ORBIT Apr 20 · 9h old	$220K	19 LOW	Graduated to Raydium, strong community	↑ 11.2x to $2.4M	✓	solscan ↗

Overall accuracy
87%
870 out of 1,000 tokens correctly predicted across all risk levels

Rug detection rate

91%

High-risk tokens (score 70+) rugged within 24h in 91% of cases

Survivor prediction

79%

Low-risk tokens (score under 35) survived past 7 days in 79% of cases

"I ran this system manually for 6 weeks before building CabalGPT into a product. I tracked every call in a spreadsheet — token, score, what happened 24 hours later. The pattern was clear: the signals that predict rugs are consistent, repeatable, and fast to detect. I just needed to automate what I was already doing by hand."

Founder, CabalGPT

Solana degen since 2021 · Built on pump.fun since launch

+$14,200

P&L during 6-week manual testing

12 rugs

Avoided by following the score

3 tokens

Entered that went 5x–11x

Backtest data represents simulated historical analysis using the CabalGPT scoring model applied retroactively to past pump.fun token launches. Past performance does not guarantee future results. All on-chain data is publicly verifiable via solscan.io. This is not financial advice. Crypto trading involves significant risk of loss.

The problem

You have 4 minutes.
Make the call.

95%

of pump.fun tokens rug or die within 24 hours. Most were detectable before launch.

4 min

Average window to enter a token before smart money positions are fully set.

500+

new tokens launch on pump.fun every single day. You're evaluating them with a gut feeling.

Your research process is the vulnerability.

You check Bubblemaps for wallet clustering. You scroll Twitter for chatter. You look at the description for red flags. You check if the GitHub is real. You estimate the MC. By the time you've done all that — the entry is gone.

Meanwhile, well-connected cabals have done this analysis in parallel across 20 tokens simultaneously, using infrastructure you don't have access to.

CabalGPT closes that gap. Six specialized AI agents, running in parallel, give you a complete verdict in under 15 seconds.

Intelligence dimensions

Every angle.
Scored.

CabalGPT doesn't give you a single number. It gives you the full picture — the same one a professional analyst would build over hours, in seconds.

Rug Score

Probability of LP pull, dev dump, or coordinated exit. Lower is safer.

Narrative Score

Strength, uniqueness, and resonance of the token's story and positioning.

Meme Power

Virality potential based on branding quality and cultural resonance signals.

Dev Trust Score

Developer credibility — wallet age, prior launches, GitHub, and track record.

Scam Probability

41%

Pattern match against known scam templates, recycled copy, and suspicious language.

AI Hype Score

Real AI utility signals vs empty buzzwords. High = genuine technical substance detected.

Holder Quality

Mix of wallet ages, hold durations, and prior trading behavior of current holders.

Cult Potential

Likelihood of forming a loyal self-sustaining holder community with identity-driven narrative.

Cabal Momentum

Evidence of coordinated community building and organized buying pressure.

Social Authenticity

Bot ratio, engagement quality, and organic growth signals across X and Telegram.

24h Survival

34%

Model probability of the token trading above entry price after 24 hours.

Social Growth

Velocity of follower and community growth across all tracked social channels.

Live example

What an analysis
looks like.

$NEURAL

7xK9...mR4p · pump.fun · 2h 14m ago

Low-med risk

Rug Score

Narrative

Meme Power

Dev Trust

✓ AI Verdict

Strong narrative with verified social traction. Dev wallet is new and unverified — position small. Twitter account 6 months old with organic engagement. No GitHub found — AI claims unverified. Enter between $60K–$90K MC. Take partial profits near $250K.

Suggested Entry

$60K–$90K

Exit Zone 1

$250K

Position Size

Small (2-3%)

Current MC

$84,200

Why CabalGPT

Built for the
speed of the chain.

⚡

15-second verdict

Six parallel agents process metadata, social signals, on-chain behavior, and narrative quality simultaneously. No waiting. No tab-switching.

◎

Trained on real outcomes

The scoring model learned from thousands of labeled pump.fun launches — rugs, survivors, and breakout tokens. It knows the patterns because it studied them.

⬡

Not just on-chain data

Every other tool only looks at wallets. CabalGPT also reads the narrative — the meme quality, the Twitter velocity, the Telegram tone, the GitHub credibility.

↗

Entry timing intelligence

Not just "is this a rug" — but "at what MC does this make sense to enter, what's the exit, and how big should your position be." Actionable, not academic.

⊛

Shareable intelligence cards

Every analysis generates a shareable score card. Post your verdict on CT before your call plays out. Build your track record. Prove your edge.

◑

Persistent dev tracking

Every developer wallet builds a history. If a dev previously rugged two tokens, that's reflected in every future token they launch — permanently.

⬡

MetaAgent

Token metadata, social links, description analysis

Active

◎

SocialAgent

X/Twitter scraping, Telegram sentiment, bot detection

Active

⊛

OnchainAgent

Holder distribution, LP lock, dev wallet behavior

Active

⟳

NarrativeAgent

Narrative classification, similarity detection, originality

Active

◑

CredibilityAgent

GitHub analysis, domain age, developer history

Active

◈

VerdictAgent

Score aggregation, final recommendation, confidence interval

Active

Six minds, running in parallel.

Most intelligence tools have a single pipeline. CabalGPT deploys a specialized agent for each dimension of analysis — each one optimized for its domain, each one cross-validating the others.

The result isn't just faster analysis. It's more accurate analysis — because agents catch what a linear pipeline misses.

Claude API for narrative analysis and scam language detection
Custom rug classifier trained on 12,000+ labeled pump.fun outcomes
Vector similarity search across historical token descriptions
Helius RPC for real-time on-chain data
Persistent dev wallet memory across all launches

FAQ

Common questions

How accurate is the rug detection? +

Our rug classifier was trained on 12,000+ labeled pump.fun token outcomes. In backtesting, it correctly identifies high-risk tokens with ~78% accuracy. We display a confidence interval on every verdict — thin data means wider uncertainty, and we tell you that explicitly. No system is 100% accurate; CabalGPT gives you a better prior, not a guarantee.

How fast is the analysis? +

Most analyses complete in 10–18 seconds. Tokens with richer social footprints (more Twitter activity, active Telegram) may take up to 25 seconds as agents process more data. We process agent outputs in parallel, so complexity doesn't compound linearly.

Can a dev manipulate my token's score? +

We've designed the scoring system to resist common manipulation vectors. Bought followers are detected via engagement ratio analysis. Coordinated buys trigger anomaly flags. Descriptions similar to known rugs score poorly even with cosmetic changes. We rate-limit scans and don't allow verified developers to scan their own tokens with premium accounts.

What chains does CabalGPT support? +

CabalGPT currently supports Solana via pump.fun only. We're building support for additional Solana launchpads next, followed by a Base integration for Ethereum-side memecoin launches. Our core architecture is chain-agnostic; the launchpad-specific work is primarily in the MetaAgent and OnchainAgent layers.

Can I pay with crypto? +

Yes. We accept USDC and SOL for all subscription tiers. We also support token-gated access — holding a sufficient amount of the CABAL token unlocks Pro access. Card payments are also available via Stripe.

Is this financial advice? +

No. CabalGPT is an intelligence tool — it surfaces signals and probabilities based on available data. Every verdict includes confidence intervals and uncertainty flags. You make the trading decisions. We help you make them with better information.

The intelligence layer
for pump.fun

We ran it on
1,000 real tokens.

You have 4 minutes.
Make the call.

Your research process is the vulnerability.

Six agents.
One verdict.

Every angle.
Scored.

What an analysis
looks like.

Built for the
speed of the chain.

Six minds, running in parallel.

Start free.
Scale with your edge.

Common questions

Stop aping blind.
Know before you go.

The intelligence layerfor pump.fun

We ran it on1,000 real tokens.

You have 4 minutes.Make the call.

Your research process is the vulnerability.

Six agents.One verdict.

Every angle.Scored.

What an analysislooks like.

Built for thespeed of the chain.

Six minds, running in parallel.

Start free.Scale with your edge.

Common questions

Stop aping blind.Know before you go.

The intelligence layer
for pump.fun

We ran it on
1,000 real tokens.

You have 4 minutes.
Make the call.

Six agents.
One verdict.

Every angle.
Scored.

What an analysis
looks like.

Built for the
speed of the chain.

Start free.
Scale with your edge.

Stop aping blind.
Know before you go.