@aegiz/sdk

On-Chain AI Threat Detection for Solana

Installation • Quick Start • Architecture • API Reference • Links

---

Aegiz is a neural network that runs inference entirely on-chain — it predicts and scores threats inside a Solana smart contract. Training happens off-chain via RLHF, with trained weights uploaded to the on-chain model. No servers. No APIs. No trust required.

This SDK lets you integrate Aegiz threat detection into your dApp, wallet, or service.

Architecture

Three independent neural networks run inference on-chain and combine into an ensemble risk score. Training is performed off-chain via RLHF (Reinforcement Learning from Human Feedback) and weight updates are uploaded to the on-chain model.

Neural Network	Features	Purpose
Wallet NN	16 → 8 → 1 (145 params)	Balance analysis, drain detection, risk streaks
Program NN	16 → 8 → 1 (145 params)	Upgrade authority, binary size, deployer reputation
Token NN	16 → 8 → 1 (145 params)	Mint/freeze authority, supply analysis, rugpull signals

435 active parameters across 3 sub-networks, with 1500 total reserved slots for future expansion.

Installation

npm install @aegiz/sdk

Peer dependencies:

npm install @solana/web3.js @coral-xyz/anchor

Quick Start

Read-Only (No Wallet Required)

import { AegizClient } from "@aegiz/sdk";
import { PublicKey } from "@solana/web3.js";

const client = new AegizClient("https://api.mainnet-beta.solana.com");

// Quick risk check — works for wallets AND programs
const result = await client.quickCheck(new PublicKey("SomeAddress..."));
console.log(result.tier);       // "low" | "moderate" | "high" | "critical"
console.log(result.riskScore);  // 0-100
console.log(result.type);       // "wallet" | "program" | "unknown"

// Get detailed wallet reputation
const rep = await client.getWalletReputation(walletPubkey);

// Get program reputation
const progRep = await client.getProgramReputation(programId);

// Get token mint analysis
const mint = await client.getTokenMintAnalysis(mintPubkey);

// Get model stats (version, accuracy, active params)
const stats = await client.getModelStats();

With Wallet (Transactions)

import { AegizTransactionClient } from "@aegiz/sdk";
import { AnchorProvider } from "@coral-xyz/anchor";

const provider = AnchorProvider.env();
const client = new AegizTransactionClient(provider);

// Predict risk for a wallet
const prediction = await client.predict(targetWallet);
console.log(prediction.riskScore);     // 0-100
console.log(prediction.drainDetected); // boolean

// Scan a program for threats
await client.scanProgram(programId, programDataPda);

// Scan a token mint for rugpull signals
await client.scanTokenMint(mintPubkey);

// Ensemble score (combines all three NNs)
await client.ensembleScore(walletRepPda, programRepPda, mintAnalysisPda);

// Community reporting & voting
await client.reportScam(scamWallet, "Drainer — stole 50 SOL", 10_000_000);
await client.reportScamProgram(scamProgram, "Fake DEX", 10_000_000);
await client.voteOnReport(blacklistEntry, true);
await client.voteOnProgramReport(blacklistEntry, false);

API Reference

AegizClient — Read-Only

Method	Description
quickCheck(address)	Quick risk lookup for any address
getWalletReputation(wallet)	Full wallet reputation data
getProgramReputation(program)	Full program reputation data
getTokenMintAnalysis(mint)	Token mint analysis (rugpull signals)
getReporterProfile(reporter)	Reporter profile and stats
getModelStats()	Model version, accuracy, active params
getBlacklistEntry(wallet)	Wallet blacklist report details
getProgramBlacklistEntry(program)	Program blacklist report details
classifyRisk(score)	Score → tier classification

AegizTransactionClient — Wallet Required

Method	Description
predict(wallet, txCount?)	Run on-chain AI inference on a wallet
scanProgram(program, data, auth?)	Scan a program for threats
scanTokenMint(mint)	Scan a token mint for rugpull signals
scanTokens(wallet, accounts?)	Scan token accounts for drainers
ensembleScore(walletRep, progRep?, mint?)	Combined ensemble risk score
reportScam(wallet, proof, stake?)	Report a wallet as scam
reportScamProgram(program, proof, stake?)	Report a program as scam
voteOnReport(entry, confirm, stake?)	Vote on a wallet report
voteOnProgramReport(entry, confirm, stake?)	Vote on a program report

PDA Helpers

import {
  getModelPda,
  getReputationPda,
  getProgramReputationPda,
  getBlacklistPda,
  getProgramBlacklistPda,
  getTreasuryPda,
  getVotePda,
  getMintAnalysisPda,
  getReporterProfilePda,
} from "@aegiz/sdk";

Network

	Value
Program ID	5QA63aoUXwHB6aUSvbzpHejKaqFCf8RafAFjhfjjRVnT
Architecture	3× (16→8→1) Ensemble
Total Parameters	435 active / 1500 reserved
Inference	On-chain (Solana BPF)
Training	Off-chain (RLHF)

Features

• 🧠 On-Chain Inference — Three 16→8→1 NNs running inference entirely on Solana
• 🎓 Off-Chain Training — RLHF-trained models with weights uploaded to the program
• 🛡️ Ensemble Scoring — Combined risk assessment from wallet, program, and token analysis
• 👥 Community Reporting — Stake-backed scam reporting with voting
• 🔍 Token Scanning — Detect honeypots, rugpulls, and drainer patterns
• 📊 Reputation Tracking — Persistent wallet and program reputation PDAs
• ⚡ Zero Trust — No servers, no APIs, everything verifiable on-chain

Links

• 🌐 Website: aegiz.io
• 🐦 Twitter: @aegizsol
• 💻 GitHub: Bytez3/aegiz-sdk
• 🔗 Explorer: View on Solana

License

MIT — see LICENSE for details.