Inside the AION pipeline

Nansen gives you the raw data. AION gives you the answer. Ten-stage pipeline, fully automated, every 4 hours. Click any stage to see the actual nansen … command it runs.

The 10-stage pipeline

~116 cr (Solana + Base) + ~31 cr (Polymarket) ≈ 147 credits / cycle, ~900 credits/day at 4h intervals
Step 1

Hot token scan

  • Pulls the top smart-money flow tokens per chain via Nansen CLI
  • Solana, Ethereum, Base in parallel
  • ~30 raw candidates per chain
5 CR/CHAIN
Step 2

Smart-money filter

  • Pulls the top buyers on each hot token
  • Keeps wallets Nansen labels as Smart Money / Fund / Whale
  • Drops retail noise immediately
5 CR + 2 CR/TOKEN
Step 3

Wallet grading

  • Runs every SM wallet through the profiler endpoint
  • Scores on PnL, win rate, trade count, consistency
  • Assigns deterministic S / A / B / C / D tier
1 CR/WALLET
Step 4

Accumulation scoring

  • Composite score from buy/sell ratio, buyer concentration (HHI), SM-buyer %, volume consistency
  • Produces one number you can rank by
  • Tunable weights (see Self-learning loop below)
0 CR (LOCAL)
Step 5

Convergence detection

  • The core unlock. Finds wallets buying ≥2 hot tokens in the same window
  • One wallet in one token = noise. Same cluster in three = syndicate
  • Maps wallet network graph for the top 5 wallets (related-wallets endpoint)
5 CR (GRAPH)
Step 6

Risk filter

  • GoPlus honeypot / rug / tax check
  • Risk tier preset: degen / balanced / conservative
  • Community blocklist voting
0 CR
Step 7

Polymarket whales

  • Same wallet+convergence pipeline applied to prediction markets
  • Multi-bet grouping detects coordinated wagers across related markets
  • Hedge detection reveals true directional conviction behind offsetting positions
  • Free Polymarket data-api fetches the full position book per whale
~31 CR/RUN
Step 8

AI reasoning

  • LLM synthesises the raw data into a verdict: strong_buy / buy / watch / avoid
  • Outputs confidence, bullish factors, bearish factors, risk flags
  • Turns 40 numbers into one sentence
LLM CALL
Step 9

Action layer

  • Paper portfolio with live P/L (DexScreener prices, free)
  • Editable SL/TP per position + auto-exit cron
  • Telegram alerts + copy-trade scaffolding
0 CR
Step 10

Self-learning

  • Curator records every snapshot to /data/evolution.json
  • Forward-price tracker fills in +24h / +48h / +7d returns
  • Scoring weights nudged when high-score tokens underperform
0 CR (LOCAL)

AION vs raw Nansen

Nansen is the best on-chain data source in crypto — it's where AION gets its input. But going to nansen.ai directly means clicking through a hundred pages by hand. Here's what AION adds on top.

CapabilityNansenAION
Raw wallet + token data
Deterministic wallet grading (S/A/B/C/D)
Accumulation composite score
Cross-token convergence join
Risk filter (honeypot / rug check)
AI verdict + reasoning
Self-evolving scoringWIP
7-investor persona panel
Multi-source research (6 APIs)
Polymarket multi-bet / hedge detection
The single most important differentiator

Cross-token convergence detection. Nansen will never tell you “these three wallets are buying Token A and Token B and Token C right now.” That join doesn't exist in their UI. One wallet in one token is noise. The same cluster showing up in three is a syndicate. This is the signal that separates alpha from coincidence.

How wallet grading works

Every wallet gets scored on five weighted factors, producing a 0–100 score that maps to a tier. The weights are tunable via the self-evolving loop.

SCORING FACTORS
Win rate30 pts

% of tracked trades that closed profitable

Realised PnL30 pts

Total $ made from closed positions over rolling window

Trade count15 pts

Sample size — more trades = more reliable signal

Consistency15 pts

Low variance across months, not one lucky hit

Nansen label10 pts

Bonus for Smart Money / Fund / Whale labels

GRADE TIERS
S90–100Elite — top 1% of tracked wallets. Systematic alpha.
A75–89High conviction. Consistently profitable across trades.
B60–74Above average. Shows strong patterns but some noise.
C40–59Mixed results. Watch list.
D0–39Inconsistent or unprofitable. Filtered out.

How accumulation scoring works

Separate from wallet grading — this scores the token, not the wallet. Answers: “Is smart money actually accumulating this, or is it just volatile noise?”

Buy/sell ratio25 pts

Smart money buying more than selling

Buyer diversity25 pts

Low Herfindahl (HHI) — spread across many wallets, not one

SM presence20 pts

% of buyers that Nansen labels as Smart Money

Volume consistency15 pts

Steady accumulation over time, not one spike

Buyer count15 pts

Raw number of distinct SM wallets

Accumulation signals

When multiple signals fire simultaneously, the conviction score increases. These are the patterns AION watches for.

Wallet convergence

Multiple high-grade wallets buying the same token within a short window.

Volume anomaly

Unusual volume spike relative to the 30-day average.

Whale accumulation

Large wallets increasing positions — high-conviction bets.

Smart money consensus

3+ S/A-grade wallets aligned on the same token — strongest confluence signal.

Risk tiers

Every token is assigned a risk tier based on liquidity, project maturity, and audit status.

TierScoreCharacteristics
Low
1–3High liquidity, established project, multiple audits.
Medium
4–6Moderate liquidity, growing community, some risk factors.
High
7–10Low liquidity, new/unaudited, high volatility.

Self-evolving scoring loop

Karpathy-style autoresearch pattern: hypothesis → experiment → evaluate → improve. Every discovery run is a live experiment.

STAGE 1
Hypothesis

Current scoring weights are loaded as the active model

STAGE 2
Experiment

Discovery run records a snapshot: every token, its score, its entry price

STAGE 3
Evaluate

Forward price checks at +24h / +48h. High-score vs low-score returns compared

STAGE 4
Improve

Weights adjusted if high-score tokens underperform. New version committed

Storage: snapshots and evaluations live in flat JSON files on the backend (data/evolution.json + data/scoring_weights.json). No database needed at this scale — a snapshot every 4 hours is ~6 records/day. The design mirrors Karpathy's markdown-wiki approach: append-only, git-friendly, human-readable.

7-investor persona panel

Every token AION surfaces gets analyzed by seven AI personas, each modeled after a legendary investor with a distinct philosophy. They pull live data from six external sources and deliver independent verdicts.

Buffett

Value investing, moats, fundamentals

Burry

Contrarian deep-value, macro shorts

Druckenmiller

Macro momentum, position sizing

Damodaran

Quantitative valuation, risk models

Wood

Disruptive innovation, long-term growth

Ackman

Activist conviction, catalyst-driven

Jhunjhunwala

Emerging market growth, high conviction

Multi-source research

Each persona pulls live data from CoinGecko (price, volume, market cap), GoPlus Security (contract risk), DexScreener (DEX liquidity), DefiLlama (TVL, protocol data), GitHub (development activity), and X/Twitter (sentiment, mentions, influencer signals). No opinions without evidence. Results are cached for 6 hours (instant on repeat lookups), and every verdict is tracked against actual forward price performance so the panel's accuracy can be measured and improved over time.

Data sources

AION aggregates intelligence from seven data providers to build a complete picture of every token and market.

Nansen CLI

Smart money netflow, wallet profiling, trade history, Polymarket screener

CoinGecko

Market data, price history, volume, market cap rankings

GoPlus Security

Honeypot detection, rug pull checks, contract risk scanning

DexScreener

Real-time DEX prices, pair data, liquidity metrics

DefiLlama

TVL tracking, protocol analytics, yield data

GitHub

Project activity, commit frequency, contributor metrics

X (Twitter) API

Sentiment analysis, mention tracking, influencer signals

Ready to get started?

Add the AION bot to Telegram and start tracking smart money in minutes.

Open Telegram bot