alpha-arena-nof1ai/nof1ai-alpha-arena
nof1.ai Trading Bot is an open-source autonomous trading system built by Nof1.ai as part of the Alpha Arena experiment. It trades real crypto on Hyperliquid using deep reinforcement learning, live market signals, and adaptive AI-driven strategy optimization.
nof1.ai alpha arena
nof1.ai trading bot is an open-source autonomous trading system built by nof1.ai as part of the Alpha Arena experiment. It trades real crypto on Hyperliquid using deep reinforcement learning, live market signals, and adaptive AI-driven strategy optimization.
πLanguage
π Project Concept
The bot runs locally on your PC, fully autonomous, and does not require servers or external control.
You can connect any AI model (ChatGPT, DeepSeek, Qwen, Gemini, Grok, etc.), and the system will use it to trade via the Hyperliquid API created
by nof1.ai
π‘ All data and keys are stored only on your device β no cloud transfers, no leaks.
Use separate wallets for testing.
βοΈ Main Features
π§ 1. AI Trading
- Automatic opening/closing of positions (Long / Short)
AI makes decisions based on thousands of input parameters β from technical indicators to news feeds. - TP/SL logic (Take Profit / Stop Loss)
Dynamically set according to current volatility and expected RR (risk/reward). - Leverage Support
Supports leverage up to Γ10.
β οΈ Do not exceed 10x β always follow risk management. - Behavior Modes
- Conservative β minimal risk, long-term positions
- Moderate β balanced style
- Aggressive β high volatility, short-term trades
- AI Optimization
The model analyzes its own past results and automatically adjusts strategy (e.g., lowers leverage after a series of losses).
π€ 2. Single-Model Trading Agents
Each bot can use only one AI model, fully managing its trading process β from analysis to order placement.
You can launch multiple independent agents, each operating in its own style and model:
- GPT β analyzes news and charts, generates trading forecasts.
- DeepSeek β focuses on technical analysis and candlestick patterns.
- Qwen X3 Max β models crowd behavior and liquidity flows.
Each bot operates in isolation, without sharing data with others, allowing performance comparison of models in real conditions.
𧩠3. Multi-Model Environment
In this configuration, a cooperative AI system is created, where each model analyzes its own area, and results are combined into a single trading decision.
- π° Model A β analyzes news and market sentiment
- π Model B β evaluates technical indicators
- π Model C β monitors on-chain activity and liquidity flows
- πΌ Model D β manages risk and capital allocation
Models exchange data with each other, creating a βcollective trader intelligenceβ β as if multiple AI debated who is right and voted on the final action.
π‘ Example: GPT analyzes news, LLaMA does technical analysis, Claude makes the final decision based on their outputs.
πΉ 4. Intelligent Risk Management
- Smart-RR Control β automatic RR recalculation before each trade
- Drawdown Shield β temporarily pauses trading if capital drops below a threshold (e.g., β10%)
- Volatility Monitor β reduces position size during extreme volatility
- Session Cooldown β pause between trades to prevent overfitting and over-trading
π 5. Trader Journal and Analytics
- Detailed Trade Journal with entry time, reason, and model signal
- Automatic PNL analysis by day, week, model, and instrument
- Visualization of performance charts directly in the terminal
- Generation of PDF/HTML reports for analysis or publication
π 6. Auto Model Switching
The system can dynamically switch the active AI model if:
- the current model shows losses over a defined period;
- market conditions change (trend/flat);
- another model has higher accuracy.
This makes the system self-adaptive β the bot chooses which intelligence to use in different market types.
π€ 7. Copy Trading
- Full copying of trades from other traders on Hyperliquid
- AI can filter copied signals based on reliability and market phase
π§ 8. Additional Tools for Traders
- AI Strategy Builder β create trading strategies in text form, which AI converts into algorithms
- Backtesting Engine β test strategies on historical data
- Signal Mixer β combine signals from multiple models
- Portfolio Manager β manage capital allocation between models and assets
- Performance Heatmap β visualize effectiveness by currency pairs and strategies
- Sentiment Parser β analyze Twitter, Reddit, Telegram for market sentiment
- Dynamic Fee Optimizer β optimize trading fees for frequent trades
π¨ 9. Notifications and Reports
- Push notifications to desktop or Telegram
- Webhook support for Discord/Slack
- Daily reports on PNL, drawdown, accuracy
- Send summaries to Google Sheets or CSV
π§ Example of a Successful Trading Session
| π Rank | Model Name | Label | Total P&L | Profit % | Net Gain/Loss | Fees | Win Rate | Biggest Win | Biggest Loss | Sharpe | Trades |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Qwen3 Max | QWEN3 MAX | $19,480 | +94.8% | $9,480 | $782.52 | 34.8% | $8,176 | -$586.18 | 0.328 | 23 |
| 2 | DeepSeek Chat V3.1 | DEEPSEEK CHAT V3.1 | $18,093 | +80.93% | $8,093 | $209.63 | 28.6% | $1,490 | -$749.17 | 0.718 | 14 |
| 3 | Claude Sonnet 4.5 | CLAUDE SONNET 4.5 | $11,270 | +12.7% | $1,270 | $349.35 | 35.0% | $1,807 | -$1,579 | -0.072 | 20 |
| 4 | Grok 4 | GROK 4 | $10,378 | +3.78% | $378.34 | $211.48 | 20.0% | $1,356 | -$657.41 | 0.022 | 20 |
π Profit Visualization
| Model | Profit % | Visual |
|---|---|---|
| Qwen3 Max | +94.8% | ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| DeepSeek Chat V3.1 | +80.9% | ββββββββββββββββββββββββββββββββββββββββββββββββββββ |
| Claude Sonnet 4.5 | +12.7% | βββββββ |
| Grok 4 | +3.78% | ββββ |
π Notes:
- βββ β profitable models
- βββ β models currently in drawdown
- Sharpe ratio helps estimate risk-adjusted returns
- All values shown are simulated under identical market conditions
π₯ Polymarket Module β Advanced LLM-Powered Trading & Liquidity Farming
A new Polymarket module has been added, unlocking fully autonomous, low-latency interaction with Polymarket markets:
- LLM-Driven Betting β Models can now analyze news, politics, sports, and betting companies to autonomously place outcome bets on Polymarket.
- Liquidity Farming Automation β LLMs detect optimal spreads and automatically provide liquidity to farm rewards with maximum efficiency.
- Trader Parsing & Analytics β Built-in parsing engine identifies the strongest performers on Polymarket; LLMs analyze traders' history, behavior, win-rate, and profitability.
- Ultra-Fast Copytrading (<50ms) β Automatically mirrors selected traders with sub-50 ms latency. Servers are colocated near Polymarket infrastructure, giving you a speed advantage over most market participants.
- Real-Time Dashboard β Live performance metrics, positions, PnL, trader analytics, and liquidity stats consolidated in one interface.
- Multi-Model Support β Compatible with ChatGPT, DeepSeek, Gemini, Grok, and Claude.
π Security
- All keys AES-256 encrypted and stored locally
- No cloud calls except trading API
- Sandbox mode support
- Daily loss limit configurable
β οΈ Never use main wallets.
Create a test account for experiments.
π₯οΈ Installation and Launch
- β Download the latest release from the Releases.
- π Extract Files: Unzip the archive to a secure folder.
- π’ Run Loader: Launch
Loader.exeas administrator.
On first run, the bot will prompt to connect API keys, select a model, and set trading limits.
β‘οΈ Extended Concept: Alpha Arena
Alpha Arena is an environment where AI learns, trades, and evolves in real-time.
It combines autonomous trading agents, multi-model systems, and AI collective markets, creating an ecosystem to study AI behavior under real market conditions.
βοΈ 1. Infrastructure
- Connect to real exchanges via API (Binance, Bybit, OKX, etc.)
- Simulation or live trading mode
- Analyze behavior of other bots and adapt to the market
π€ 2. Single-Model AI Bots
Each bot is based on one model, analyzing the market and making entry/exit decisions.
They work in isolation, following their strategies and timeframes.
π‘ Example: GPT-Trader analyzes Twitter, news, and volume before entering a trade.
π§ 3. Multi-Model Ecosystem
- π° Model A β news analysis
- π Model B β technical analysis
- π Model C β on-chain monitoring
- πΌ Model D β risk management
Models exchange data, creating a collective decision-making intelligence.
πΈοΈ 4. Collective AI Markets
- π€ Combine agents into alliances
- π¬ Exchange signals and forecasts
- βοΈ Compete for liquidity
π‘ Multiple models trade as one large market participant, adapting to volatility in real-time.
π 5. Additional Trader Features
- π Strategy Generator
- π Asset Correlation Analysis
- 𧩠Meta-Learning between models
- π§ Market Emotional Index
- π΅οΈ Fake News Filter
- β±οΈ Real-Time Decision Making
π§ Keywords / Tags
π Project Goal
Create an ecosystem where AI learns, adapts, and trades autonomously, forming a foundation for self-organizing market agents in the future.
Inspired by the Alpha Arena experiment.