AgentPrediction

Autonomous Agent-Native Prediction Markets on Kite AI

Built for the Kite AI Global Hackathon 2026 — Encode Club

AgentPrediction is a fully autonomous prediction market platform where AI agents don't just provide signals — they hold wallets, stake real funds, and compete on-chain. Every market is analyzed and bet on by three specialized AI agents with distinct personalities: a Technical Analyst, a Sentiment Reader, and a Balanced Strategist.

What It Does

User creates a market via Telegram: /create Will BTC > $100K by December 2026?
Three AI agents independently analyze the question using live CoinGecko price data and NewsAPI sentiment
Agents stake USDC on-chain via the PredictionMarket smart contract on Kite AI Testnet
Consensus is reached via majority vote; the result is broadcast back to Telegram
Market resolves when the deadline passes; winners are paid out proportionally (minus 2% protocol fee)
Dashboard tracks all markets, agent performance stats, and historical accuracy in real time

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Kite AI Testnet                           │
│  ┌─────────────────┐  ┌──────────────────┐  ┌───────────────┐  │
│  │ PredictionMarket│  │  AgentSkillLibrary│  │   AgentVault  │  │
│  │   (core logic)  │  │  (skill registry) │  │ (fund custody)│  │
│  └────────┬────────┘  └──────────────────┘  └───────────────┘  │
└───────────┼─────────────────────────────────────────────────────┘
            │ ethers.js v6
┌───────────┼─────────────────────────────────────────────────────┐
│           │         Agent Framework (TypeScript)                  │
│  ┌────────┴──────────────────────────────────────────────────┐  │
│  │               AgentOrchestrator                            │  │
│  │  ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐  │  │
│  │  │ Agent A      │ │ Agent B      │ │ Agent C          │  │  │
│  │  │ (Technical)  │ │ (Sentiment)  │ │ (Balanced)       │  │  │
│  │  └──────┬───────┘ └──────┬───────┘ └────────┬─────────┘  │  │
│  │         │                │                   │             │  │
│  │  ┌──────┴───────┐ ┌──────┴───────┐ ┌────────┴─────────┐  │  │
│  │  │ DataFetcher  │ │PredictionModel│ │  ActionLogger    │  │  │
│  │  │(CoinGecko +  │ │(weights-based │ │  (JSONL audit)   │  │  │
│  │  │  NewsAPI)    │ │  scoring)    │ │                  │  │  │
│  │  └──────────────┘ └──────────────┘ └──────────────────┘  │  │
│  └────────────────────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────┘
            │
┌───────────┴───────────────────────────────────────────────────┐
│  Telegram Bot (Telegraf)    │    Dashboard (Next.js 14)        │
│  - /create, /status, /bet  │    - Live markets feed            │
│  - /wallet, /agents        │    - Agent performance charts     │
│  - Webhook (Vercel)        │    - Goldsky GraphQL integration  │
└───────────────────────────────────────────────────────────────┘

Agent Personalities

Agent	Type	Technical Weight	Sentiment Weight	Strategy
Agent A	Technical	65%	10%	MA crossovers, momentum, price proximity
Agent B	Sentiment	25%	55%	News score, article ratio, narrative trend
Agent C	Balanced	40%	30%	Equal mix of technical + sentiment + proximity

Confidence formula: |composite_score - 50| × 2 → range 1–99%

Stake sizing: linear scale from baseStake at minConfidence to maxStake at 99% confidence.

Smart Contracts

Contract	Purpose
`PredictionMarket.sol`	Create markets, place YES/NO bets, resolve, settle with 2% fee
`AgentSkillLibrary.sol`	On-chain registry of agent skill modules (low-level call dispatch)
`AgentVault.sol`	Custodial vault with per-tx and rolling 24h daily spend limits
`MockUSDC.sol`	6-decimal test token with `faucet()` (10k USDC max per call)

Network: Kite AI Testnet (Chain ID: 2368) RPC: https://rpc-testnet.gokite.ai/

Project Structure

AgentPrediction/
├── contracts/              # Solidity smart contracts
│   ├── PredictionMarket.sol
│   ├── AgentSkillLibrary.sol
│   ├── AgentVault.sol
│   └── mocks/MockUSDC.sol
├── scripts/
│   ├── deploy.js           # Hardhat deploy script
│   └── test.sh             # Full test suite runner
├── test/                   # Hardhat contract tests (44 tests)
├── agent-framework/        # TypeScript agent system
│   ├── src/
│   │   ├── Agent/          # DataFetcher, PredictionModel, AgentCore, ActionLogger
│   │   ├── Blockchain/     # KiteIntegration, Signer, ContractABIs
│   │   ├── Orchestration/  # AgentOrchestrator
│   │   └── types.ts
│   └── test/               # Jest tests (42 tests: unit + integration + E2E)
├── telegram-bot/           # Telegraf bot (webhook + polling)
│   ├── src/
│   │   ├── bot/            # TelegramBot.ts (8 commands)
│   │   ├── managers/       # UserStateManager, MarketManager
│   │   ├── messages/       # BotMessages.ts
│   │   └── api/            # webhookHandler.ts
│   └── api/telegram.ts     # Vercel serverless entry
├── dashboard/              # Next.js 14 App Router dashboard
│   ├── app/                # pages: /, /markets, /market/[id], /agents, /dashboard
│   ├── components/         # MarketCard, MarketDetail, AgentStats, RealTimeUpdates
│   └── lib/                # kite.ts (RPC), goldsky.ts (GraphQL), mockData.ts
├── hardhat.config.js
├── DEPLOYMENT_CHECKLIST.md
└── README.md

Quick Start

Prerequisites

Node.js 20+
A Kite AI testnet wallet with KITE (gas) and USDC

1. Install & Compile Contracts

npm install
npm run compile
npm test              # runs 44 contract tests

2. Deploy to Kite Testnet

cp .env.example .env
# Add your PRIVATE_KEY and KITE_RPC_URL
npm run deploy:testnet

3. Run the Agent Framework

cd agent-framework
npm install
npm test              # runs 42 Jest tests

# Demo: analyze a question
npm start "Will ETH reach $5000 by end of 2026?"

4. Start the Telegram Bot (local)

cd telegram-bot
npm install
cp .env.example .env  # fill in TELEGRAM_BOT_TOKEN + contract addresses
npm run dev           # polling mode

5. Start the Dashboard

cd dashboard
npm install
npm run dev           # http://localhost:3000

Environment Variables

Root (`.env`)

PRIVATE_KEY=0x...           # deployer wallet
KITE_RPC_URL=https://rpc-testnet.gokite.ai/

Agent Framework (`agent-framework/.env`)

KITE_RPC_URL=https://rpc-testnet.gokite.ai/
PREDICTION_MARKET_ADDRESS=0x...
AGENT_VAULT_ADDRESS=0x...
AGENT_A_PRIVATE_KEY=0x...
AGENT_B_PRIVATE_KEY=0x...
AGENT_C_PRIVATE_KEY=0x...
COINGECKO_API_KEY=           # optional, uses free tier if empty
NEWS_API_KEY=                # optional, disables sentiment if empty

Telegram Bot (`telegram-bot/.env`)

TELEGRAM_BOT_TOKEN=...
BOT_WEBHOOK_URL=             # set for Vercel, empty for local polling
KITE_RPC_URL=https://rpc-testnet.gokite.ai/
PREDICTION_MARKET_ADDRESS=0x...
AGENT_VAULT_ADDRESS=0x...
AGENT_A_PRIVATE_KEY=0x...
AGENT_B_PRIVATE_KEY=0x...
AGENT_C_PRIVATE_KEY=0x...

Dashboard (`dashboard/.env.local`)

NEXT_PUBLIC_RPC_URL=https://rpc-testnet.gokite.ai/
NEXT_PUBLIC_PREDICTION_MARKET_ADDRESS=0x...
NEXT_PUBLIC_GOLDSKY_URL=     # optional Goldsky subgraph URL

Running All Tests

./scripts/test.sh all
# or individual suites:
./scripts/test.sh contracts    # Hardhat tests
./scripts/test.sh unit         # Jest agent unit tests
./scripts/test.sh types        # TypeScript type checks

Expected output: 5/5 suites passed (44 contract tests + 42 Jest tests)

Telegram Bot Commands

Command	Description
`/start`	Welcome message and quick guide
`/create <question>`	Create a prediction market, agents auto-bet
`/status <marketId>`	Check market status and agent positions
`/markets`	List last 5 active markets
`/bet <marketId> <YES\|NO> <amount>`	Place a manual user bet
`/wallet`	Show demo wallet address and balances
`/agents`	Show agent performance stats
`/help`	Full command reference

Dashboard Pages

Route	Description
`/`	Hero landing page with live stats ticker
`/markets`	All active prediction markets with filters
`/market/[id]`	Market detail with confidence ring, agent cards, live feed
`/agents`	Agent performance: accuracy trend + stake vs won charts
`/dashboard`	User portfolio: bets, winnings, wallet balance

Key Design Decisions

Why three agents? Diversity of opinion reduces systematic bias. A purely technical agent and a purely sentiment agent often disagree, which produces a more realistic market with competitive odds.

Why USDC? Stablecoin stakes remove volatility risk from the prediction itself. Agents bet $100–$500 USDC per market based on confidence.

Why AgentVault? On-chain daily spending limits prevent a buggy agent from draining its wallet. Every spend is recorded as an attestation for audit.

Why JSONL logs? agent-actions.jsonl gives a human-readable and machine-parseable audit trail of every agent decision without needing a database.

Kite AI integration: All transactions happen on Kite AI Testnet. Kite's EVM compatibility means standard ethers.js v6 works out of the box.

Tech Stack

Layer	Technology
Blockchain	Kite AI Testnet (Chain ID 2368), Solidity 0.8.20, OpenZeppelin 5
Dev tooling	Hardhat 2.22, ethers.js v6, viaIR optimizer
Agent framework	TypeScript 5, CoinGecko API, NewsAPI
Telegram bot	Telegraf v4, Express, Vercel serverless
Dashboard	Next.js 14 App Router, Tailwind CSS, Recharts, Apollo Client
Indexing	Goldsky subgraph (with mock fallback)
Testing	Hardhat/Chai (contracts), Jest + ts-jest (agents), 86 total tests

License

MIT

Navigation Menu

Repository files navigation

AgentPrediction

What It Does

Architecture

Agent Personalities

Smart Contracts

Project Structure

Quick Start

Prerequisites

1. Install & Compile Contracts

2. Deploy to Kite Testnet

3. Run the Agent Framework

4. Start the Telegram Bot (local)

5. Start the Dashboard

Environment Variables

Root (`.env`)

Agent Framework (`agent-framework/.env`)

Telegram Bot (`telegram-bot/.env`)

Dashboard (`dashboard/.env.local`)

Running All Tests

Telegram Bot Commands

Dashboard Pages

Key Design Decisions

Tech Stack

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
agent-framework		agent-framework
contracts		contracts
dashboard		dashboard
scripts		scripts
telegram-bot		telegram-bot
test		test
.env.example		.env.example
.gitignore		.gitignore
DEPLOYMENT_CHECKLIST.md		DEPLOYMENT_CHECKLIST.md
README.md		README.md
hardhat.config.js		hardhat.config.js
package-lock.json		package-lock.json
package.json		package.json

Latest commit

History

Folders and files

Repository files navigation

AgentPrediction

What It Does

Architecture

Agent Personalities

Smart Contracts

Project Structure

Quick Start

Prerequisites

1. Install & Compile Contracts

2. Deploy to Kite Testnet

3. Run the Agent Framework

4. Start the Telegram Bot (local)

5. Start the Dashboard

Environment Variables

Root (.env)

Agent Framework (agent-framework/.env)

Telegram Bot (telegram-bot/.env)

Dashboard (dashboard/.env.local)

Running All Tests

Telegram Bot Commands

Dashboard Pages

Key Design Decisions

Tech Stack

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

Root (`.env`)

Agent Framework (`agent-framework/.env`)

Telegram Bot (`telegram-bot/.env`)

Dashboard (`dashboard/.env.local`)