VerdictSwarm
APIScan DemoProofPortfolioPricingDocsCommunityFAQ

How It Works

The 6-agent swarm, scoring system, and everything you need to understand your scan results.

The 6 Agents

VerdictSwarm deploys 6 independent AI agents, each analyzing a different dimension of risk. No single model sees the whole picture — that's the point.

AgentWhat It AnalyzesRole
</> TechnicianOn-chain data: holders, liquidity, concentration, transaction patternsScout
SecurityContract safety: mint/freeze authority, LP locks, honeypot detection, audit statusScout
TokenomicsSupply distribution, unlock schedules, FDV vs market cap, dilution riskInvestigator
SocialX/Twitter sentiment, whale activity, fake engagement, breaking newsInvestigator
MacroMarket conditions, BTC dominance, alt season signals, sector rotationInvestigator
Devil's AdvocateAdversarial challenge — finds the worst-case scenario for every tokenChallenger

How the Swarm Works

Independent Scoring

On deep Pro scans, each agent scores the token independently on a 0–10 scale before seeing any other agent's result. No groupthink.

Adversarial Cross-Check

When agents disagree significantly, deep scans run extra cross-examination rounds — no manual escalation needed.

4-Phase Cross-Exam Loop

Challenge → Defense → Rebuttal → Resolution. Agents must justify position changes with evidence.

Weighted Consensus

Final score is a weighted consensus across all angles. Some cross-examinations end in Split Verdict — both positions are shown to you.

Why a swarm? A single AI model can be confidently wrong. Deep scans use six specialized agents to pick the token apart from different angles, plus a devil's advocate hunting for what the others missed.

Score & Grade Scale

The backend action — AVOID, CAUTION, or CLEAR — is the authoritative pre-trade signal. The score and grade summarize supporting analysis; they do not override the action.

ScoreGrade
90–100A+
80–89A
70–79B
60–69C
40–59D
0–39F

Common Questions

What do AVOID, CAUTION, and CLEAR mean?

AVOID means the scan found a blocking risk signal. CAUTION means material concerns or uncertainty need more verification. CLEAR means no safety-policy blocker was identified in the available data; it is not a buy recommendation. This action is authoritative, while the score, grade, and evidence provide supporting detail.

Why does the Devil's Advocate often score low?

That's its job. The Devil's Advocate is designed to find the worst-case scenario so the other agents must defend their positions. A token that survives the Devil's Advocate is stronger for it — its low score is the pressure that makes the rest of the swarm justify their optimism.

Why did a popular token get a low score?

The swarm reads the chain, not the hype. If 70% of supply is in 10 wallets, that's a risk regardless of brand name, Twitter following, or how many influencers have mentioned it.

Is this financial advice?

No. VerdictSwarm provides automated risk assessments based on on-chain and market data. Results are for informational purposes only. Always do your own research (DYOR) before making any financial decision.

How fresh is the data?

Scans use live on-chain data and real-time X/Twitter search at the moment of the scan. Cached results are up to 1 hour old — the timestamp is shown on every report.

Frequently Asked Questions

What is VerdictSwarm?+
An AI-powered token security platform supporting Solana, Base, and Virtuals Protocol that evaluates token risk before a wallet, agent, or trading bot moves money. It combines deterministic checks, live market data, on-chain signals, and deep multi-agent analysis for higher tiers.
How is this different from other scanners?+
Most scanners use a single model or static rules. VerdictSwarm separates fast machine verdicts from deep swarm analysis, then returns an explicit avoid, caution, or clear action through POST /v2/verdict. Deep scans add cross-checking and adversarial review when the signal mix is ambiguous.
Is the free tier actually free?+
Yes. Anonymous users get the free fast lane, currently 3 fast scans per day. Connect a wallet for the current free deep-scan allowance shown in the app, within the 5 scans/day free cap.
How do I upgrade to higher tiers?+
Pro is $19.99/month or $15.99/month billed annually. It unlocks the full 6-agent swarm, dynamic expertise weighting, bidirectional Devil's Advocate, full reasoning with every angle shown, 30-day scan history, portfolio workflow, and 50 scans/day.
What AI models power the swarm?+
The swarm is model-agnostic and provider-diverse. Specific provider routing can change as quality and latency improve, so the product contract is the verdict, evidence, and action returned by the API rather than a fixed provider-to-agent mapping.
Can my AI agent/bot use VerdictSwarm?+
Yes. Call POST /v2/verdict directly with x402 USDC micropayments or a prepaid API key. Current machine-buyer prices are triage $0.001, fast $0.01, and deep $0.025; cache hits are 20% of the fresh price. Fast verdicts are about 2.5s p95, while deep scans typically take 30-45s.
What chains are supported?+
Solana, Base, and Virtuals Protocol are supported today. Solana tokens use Solana data providers, and Base tokens, including Virtuals Protocol AI agents, use EVM contract and market-data checks. Additional chains are not live yet.
How accurate is it?+
Verdict quality improves with data coverage and scan depth. The free fast lane is built for quick avoid, caution, or clear checks. Deep Pro scans add the full 6-agent swarm, Devil's Advocate review, and richer reasoning for ambiguous or higher-stakes tokens.
Where are verdict receipts stored?+
No public immutable verdict-storage product is live today. VerdictSwarm currently exposes API responses, report pages, and proof-style receipt surfaces for verification.
Do you keep up with new AI models?+
Yes. The swarm architecture is model-agnostic, so we can change provider routing, prompts, and analysis strategies as better options become available without changing the API contract your integration depends on.
Is your core technology proprietary?+
Yes. VerdictSwarm's adversarial consensus engine, prompt architecture, and scoring methodology are proprietary technology. Think of it like PageRank for Google — continuously tuned, continuously upgraded, and central to why our verdict quality compounds over time.

Ready to see it in action?

Scan a Token
View full documentation

A Sentien Labs Product · VerdictSwarm

Live on VirtualsDocsAPI & AgentsTermsPrivacyRefundsGitHubX

Analysis only — not financial advice. DYOR.

OPINION ONLY - NOT FINANCIAL ADVICE
VerdictSwarm provides automated risk assessments based on public on-chain data. All “Verdicts” and “Risk Ratings” are opinions generated by AI models, not statements of fact. An “AVOID” verdict does not accuse anyone of crimes. A “CLEAR” verdict does not guarantee safety. Conduct your own research. We accept no liability for losses.