ERC-7857 iNFT · 0G Network · Uniswap V3

PayAgent

Your AI financial agent, on-chain.

Autonomous paycheck allocation → real token swaps → learns from your feedback

The Problem

"Every month,
the same ritual."

You manually split your paycheck across bills, savings, and investments. You forget your own rules. You make emotional decisions when markets move.

The Solution

01 Set bills and goals once.

02 Say how much you earned.

03 Agent allocates + executes real swaps.

04 Override it? It learns.

No retraining. In-context learning from override history.

Architecture

Telegram Bot

User interface

→

0G Compute

LLM inference

→

Agent Core

Allocation logic

→

Uniswap V3

Real swaps on Sepolia

0G Storage

Decentralized persistence — config, bills, goals, decisions

ERC-7857 iNFT

On-chain data hash anchor — verifiable ownership

Swaps

4 mock tokens, 3 Uniswap V3 pools on official Sepolia factory. Real on-chain transactions.

Quotes

3-tier fallback: QuoterV2 → staticCall → pool slot0() sqrtPriceX96 math.

Failure

Swap fails? Funds stay as USDC. Never crashes.

Paycheck #1 — Approve

$5,000 — Approved

Item	Amount	Token	Action
Rent	$1,500	USDC	Reserve
Utilities	$200	USDC	Reserve
Internet	$80	USDC	Reserve
ETH Savings (40%)	$1,288	WETH	Swap →
Stablecoin (30%)	$966	DAI	Swap →
Emergency (rest)	$966	USDC	—

2 swaps executed

1,288 USDC → 0.639 WETH
966 USDC → 963 DAI

on-chain

Real Sepolia tx hashes.
Decision → 0G Storage.
iNFT hashes updated.

Paycheck #2 — Override

$5,000 — Overridden

AI proposed 40% ETH again. User disagrees:

"Market is volatile. Reduce ETH to 15%, put more in DAI."

ETH

40% → 15%

DAI

30% → 55%

// decision record
{
  outcome: "overridden",
  reason:  "Market is volatile,
           reduce ETH exposure",
  originalPlan: { /* 40% ETH */ },
  finalPlan:    { /* 15% ETH */ },
  txHashes: ["0xa3f...", "0xb7c..."]
}
// saved to 0G Storage + iNFT updated

This is the learning signal

Paycheck #3

The agent adapts.

ETH allocation across three months:

Month 1 — approved40%

$1,288

Month 2 — overridden15%

$483

Month 3 — agent adapted25%

$805

Full decision history — including overrides with reasons — passed as LLM context. Pure in-context learning.

3

Months processed

6

On-chain swaps

1

Override learned from

Built With

0G Compute

LLM inference (deepseek-chat)

0G Storage

Decentralized data layer

ERC-7857 iNFT

On-chain verifiable agent

Uniswap V3

Real swaps on Sepolia

grammY

Telegram bot framework

ethers.js v6

Contract interaction

TypeScript

pnpm monorepo

Hardhat

Contract deployment

Portable. Verifiable.
Learns from you.

Your AI financial agent lives as an NFT.
Its data is decentralized. Its decisions are on-chain.
It gets better every month.

Thank you.