Fenix AI 交易机器人

Prerequisites

Install dependencies

pip install -e ".[dev,llm,vision,monitoring]" python -m playwright install chromium plotly_get_chrome -y

Pull required Ollama models

ollama pull qwen3:8b ```

Install dev dependencies

pip install -e ".[dev]"

Installation

```bash

Docker

```bash cp .env.example .env chmod 600 .env

Development Setup

```bash

Install pre-commit hooks

pre-commit install

Example: Agent retrieves relevant context from ReasoningBank

context = reasoning_bank.get_relevant_context( agent_name="technical_analyst", current_prompt=market_analysis_prompt, limit=3 )

🚀 Quick Start

Optional Services

• Binance Account - For live/testnet trading
• Groq API Key - For cloud LLM inference
• HuggingFace Token - For HF Inference API
• Playwright - For TradingView chart capture

Create and activate virtual environment

python -m venv .venv source .venv/bin/activate # Linux/Mac

Configure environment

cp .env.example .env

Edit .env with your API keys

CLI Options

python run_fenix.py --help

python run_fenix.py                      # Paper trading (default)
python run_fenix.py --symbol ETHUSDT     # Different symbol
python run_fenix.py --timeframe 5m       # Different timeframe
python run_fenix.py --mode paper --mainnet-data  # Mainnet public data, simulated orders
python run_fenix.py --no-visual          # Disable visual agent
python run_fenix.py --mode live --allow-live  # Live trading (⚠️ real money)

For isolated multi-timeframe paper research, use scripts/paper_experiment_harness.sh. It can run against Testnet or Mainnet public data, separates each candidate's database and memory, strips Binance credentials, and includes an aggregate-only inspector. See docs/PAPER_EXPERIMENT_HARNESS.md.

---

📊 Configuration

Main Configuration (`config/fenix.yaml`)

trading:
  symbol: BTCUSDT
  timeframe: 15m
  max_risk_per_trade: 0.02
  
agents:
  enable_technical: true
  enable_qabba: true
  enable_visual: true  # Requires vision model
  enable_sentiment: true  # Requires news APIs
  technical_weight: 0.30
  qabba_weight: 0.30
  consensus_threshold: 0.65

LLM Providers (`config/llm_providers.yaml`)

active_profile: "all_local"  # Options: all_local, mixed_providers, mlx_optimized, all_cloud

all_local:
  technical:
    provider_type: "ollama_local"
    model_name: "qwen3:8b"
    temperature: 0.1

Environment Variables

---

Terminal 1: Start the backend with API

python run_fenix.py --api

API + Redis

docker compose up -d --build

API + Redis + Prometheus + Grafana

docker compose --profile monitoring up -d --build ```

Docker defaults to Python 3.12, publishes the API only on 127.0.0.1:8001, and keeps Redis internal to the Compose network. For local (non-Docker) runs, the API defaults to 127.0.0.1:8000; override it with FENIX_API_PORT or --port. For Docker Compose, change the published host port with FENIX_API_HOST_PORT while the container remains on port 8000.

---

👁️ Vision Pipeline Overhaul

New LLM Integrations

FenixAI supports configurable local and cloud inference providers. No model is presented as profitable or universally suitable. Evaluate provider and prompt changes with isolated paper runs using your own data, costs, and risk limits.

See NanoFenix HTF v2.5 changes and the release checklist.

---

Notable security and developer workflow improvements

• API binds to 127.0.0.1 by default to avoid accidental public exposure. To bind to all interfaces intentionally, set ALLOW_EXPOSE_API=true.
• Demo accounts are not seeded by default; set CREATE_DEMO_USERS=true for local development.
• DEFAULT_DEMO_PASSWORD and DEFAULT_ADMIN_PASSWORD may be used for local testing; avoid using them in production.
• We added DEVELOPMENT.md and RELEASE_CHECKLIST.md to help developers follow the release process and avoid secrets leaks.
• Archived internal reports are now in docs/archives/reports/ to reduce root clutter.

---

🔄 Agent Workflow

┌─────────────────────────────────────────────────────────────────────────────┐
│                              FENIX AI v2.5 RC                               │
├─────────────────────────────────────────────────────────────────────────────┤
│                                                                             │
│  ┌─────────────┐    ┌──────────────────────────────────────────────────┐    │
│  │   Frontend  │◄──►│              FastAPI + Socket.IO                 │    │
│  │  React/Vite │    │                  (Real-time)                     │    │
│  └─────────────┘    └────────────────────┬─────────────────────────────┘    │
│                                          │                                  │
│  ┌───────────────────────────────────────▼──────────────────────────────┐   │
│  │                      TRADING ENGINE                                  │   │
│  │  ┌─────────────────────────────────────────────────────────────────┐ │   │
│  │  │                 LangGraph Orchestrator                          │ │   │
│  │  │                   (State Machine)                               │ │   │
│  │  └─────────────────────────────────────────────────────────────────┘ │   │
│  │           │              │              │              │             │   │
│  │     ┌─────▼─────┐  ┌─────▼─────┐  ┌─────▼─────┐  ┌─────▼─────┐       │   │
│  │     │ Technical │  │  Visual   │  │ Sentiment │  │   QABBA   │       │   │
│  │     │  Agent    │  │  Agent    │  │  Agent    │  │  Agent    │       │   │
│  │     └─────┬─────┘  └─────┬─────┘  └─────┬─────┘  └─────┬─────┘       │   │
│  │           │              │              │              │             │   │
│  │     ┌─────▼──────────────▼──────────────▼──────────────▼─────┐       │   │
│  │     │              Decision Agent + Risk Manager             │       │   │
│  │     │           (Dynamic Weighting + LLM-as-Judge)           │       │   │
│  │     └────────────────────────┬───────────────────────────────┘       │   │
│  └──────────────────────────────┼───────────────────────────────────────┘   │
│                                 │                                           │
│  ┌──────────────────────────────▼───────────────────────────────────────┐   │
│  │                         MEMORY LAYER                                 │   │
│  │  ┌─────────────────┐  ┌─────────────────┐  ┌─────────────────────┐   │   │
│  │  │  ReasoningBank  │  │  Trade Memory   │  │   LLM-as-Judge      │   │   │
│  │  │ (Semantic Search)│ │   (History)     │  │  (Self-Evaluation)  │   │   │
│  │  └─────────────────┘  └─────────────────┘  └─────────────────────┘   │   │
│  └──────────────────────────────────────────────────────────────────────┘   │
│                                                                             │
│  ┌──────────────────────────────────────────────────────────────────────┐   │
│  │                        EXECUTION LAYER                               │   │
│  │  ┌─────────────────┐  ┌─────────────────┐  ┌─────────────────────┐   │   │
│  │  │  Binance Client │  │  Order Executor │  │   Market Data       │   │   │
│  │  │ (REST + WS)     │  │  (Paper/Live)   │  │   (Real-time)       │   │   │
│  │  └─────────────────┘  └─────────────────┘  └─────────────────────┘   │   │
│  └──────────────────────────────────────────────────────────────────────┘   │
│                                                                             │
└─────────────────────────────────────────────────────────────────────────────┘

Run integration tests

pytest tests/test_integration.py -v