AI-powered quantitative trading platform with multi-chain autonomous agents.
Binalyst combines real-time market signals, natural-language strategy authoring, a regime-aware adaptive engine, and on-chain execution into a unified agent dashboard — running live on multi chain.
- AI Trading Agent — autonomous loop that perceives market conditions, decides on a strategy, and executes trades on BNB Chain every 2 minutes
- Natural Language Strategy Builder — write strategies in plain English ("when RSI drops below 30 and MACD crosses bullish, buy 5%") and the engine converts them to executable rules
- Regime-Aware Adaptation — detects TRENDING / RANGING / FLAT market states and switches strategy mode automatically
- Full Backtester — validates strategies on real Binance OHLCV candles with no lookahead bias; returns Sharpe ratio, max drawdown, win rate, equity curve
- Sui Agent — Move-policy-gated agent wallet on Sui with Walrus activity logging and DeepBook order support
- ERC-8004 Identity — every agent mints an onchain identity NFT on Celo and Mantle via the ERC-8004 registry
| Layer | Tools |
|---|---|
| Framework | Next.js 15, TypeScript, Tailwind CSS |
| AI | Claude (Anthropic) via claude.ts |
| Blockchain | Ethers.js v6 (BNB SDK) |
| Market Data | Binance public API, Bybit, CMC Fear & Greed, Bitget Skill Hub |
| State | Zustand (independent stores per chain) |
| Auth / DB | Supabase, NextAuth |
| On-chain storage | Walrus (Sui activity logs) |
| Deployment | Vercel |
┌─────────────────────────────────────────────────────┐
│ Binalyst Agent │
│ │
│ PERCEPTION │
│ ├── Bitget Skill Hub: 23 technical indicators │
│ │ (RSI, MACD, BB, ADX, EMA, OBV, ATR, VWAP…) │
│ ├── CMC: Fear & Greed index, trending tokens │
│ └── Binance: OHLCV klines (public API) │
│ │
│ DECISION │
│ ├── AdaptiveStrategy: regime → mode → StrategyRules │
│ ├── NL parser: plain English → StrategyRule[] │
│ └── evaluateRules(): condition matching per bar │
│ │
│ EXECUTION │
│ ├── agentLoop.ts: ticks every 2 minutes │
│ ├── TWAK client: BEP-20 swap execution on BSC │
│ └── dryRun mode: simulate without signing │
│ │
│ RISK MANAGEMENT │
│ ├── maxDrawdownPct guard (hard stop < 30%) │
│ ├── maxPerTradePct limit │
│ ├── maxDailyTrades cap │
│ └── slippage + competition compliance checks │
│ │
│ VALIDATION │
│ ├── Backtester: real Binance OHLCV, 200-bar window │
│ ├── Metrics: return, Sharpe, drawdown, win rate │
│ └── Equity curve + full trade log │
└─────────────────────────────────────────────────────┘
- Node.js 18+
- A funded wallet for whichever chains you want to use
- API keys (see Environment Variables below)
git clone https://github.com/davife2025/Binalyst.git
cd Binalyst
npm install
cp .env.celo.example .env.local # or .env.mantle.example — see below
npm run devOpen http://localhost:3000.
The .env.celo.example and .env.mantle.example files in the repo root document every variable. Key ones:
| Variable | Purpose | Required |
|---|---|---|
ANTHROPIC_API_KEY |
NL strategy parsing, AI chat | Yes |
BITGET_API_KEY |
Bitget Skill Hub routing (optional) | No — falls back to local computation |
NEXT_PUBLIC_SUPABASE_URL / NEXT_PUBLIC_SUPABASE_ANON_KEY |
Auth & persistence | For auth features |
Never commit private keys. All agent wallets are self-custodial; keys stay in your environment and are never sent to any third party.
- 23-indicator technical analysis via
lib/skills/bitget-technicals.ts - TWAK client for BEP-20 swap execution
- Adaptive strategy engine with TRENDING / RANGING / FLAT regimes
- Full backtester at
/api/backtest
- Signals tab — live Fear & Greed index + 23 technical signals for any pair
- Strategy Builder — load the BTC Adaptive template, click Parse
- Backtest tab — run BTC/1h/90d/$10k starting capital → see equity curve and metrics
- Agent tab — start the agent, watch live regime detection update every 2 minutes
No lookahead bias in the backtester. Rules are evaluated on bar i but fill at candles[i+1].open. The strategy never sees future data.
Regime-aware adaptation. Rather than a fixed strategy, the engine reads ADX + EMA stack + ATR to detect market state and switches behaviour automatically. Ranging markets use RSI/BB mean-reversion; trending markets use EMA-cross + MACD; flat markets hold cash.
Bitget skill fallback. All 23 indicators are computed locally from Binance public klines. When BITGET_API_KEY is set, routing switches to the Bitget Skill Hub REST endpoint. Zero dependency on a paid API for the demo.
Fully parallel chain modules. The Celo and Mantle agents share no code with the BNB trading agent. Each chain has its own store, client, loop, and API routes — all run side-by-side without interference.
On-chain-only agent identity. ERC-8004 registration files are stored as data:application/json;base64,... URIs — no IPFS or external hosting dependency.
app/
api/
agent/ # BNB agent endpoints (loop, execute, strategy, portfolio)
backtest/ # Backtester route
ai/chat/ # Claude chat route
components/
tabs/ # AgentTab, BacktestTab, CeloAgentTab, MantleAgentTab, etc.
agent/ # PnLChart, PortfolioBreakdown, FearGreedGauge, DrawdownGauge
lib/
skills/ # bitget-technicals, cmc, byreal, web3, square
signalEngine.ts # Regime detection + technical signal evaluation
playbook.ts # Bitget Playbook JSON export
move/
sources/agent_policy.move # Sui Move contract
MIT