Modular AI agent platform with multi-provider LLM support.
📚 This is a demo project for learning and exploration purposes.
| Feature | Description |
|---|---|
| 🔌 Plug & Play LLMs | Switch between Gemini, OpenRouter, or any OpenAI-compatible API |
| 🧩 Modular Agents | Each agent is independent — add your own in minutes |
| 🛠️ Real Tool Execution | Agents don't just chat — they do things: search, code, analyze |
| 📊 Live Crypto Data | Binance API integration with real-time prices, orderbook, trades |
| 🐳 Docker Sandbox | Code execution in isolated containers — safe and reproducible |
| ⚡ Native Tool Calling | Gemini 2.5 Pro uses native function calling for reliable execution |
Intelligent conversational agent with web capabilities
- Smart Search — Multi-source web search with result aggregation
- Page Reading — Extract and summarize content from any URL
- Context Awareness — Maintains conversation history and session state
Full-featured coding assistant that actually executes code
- File Operations — Create, read, list files in isolated workspace
- Code Execution — Run Python in sandboxed Docker environment
- Multi-File Projects — Import between files, build complete projects
- Native Tool Calling — Gemini 2.5 Pro uses native function calling
- Tool Support Indicator — ✅/
⚠️ shows which models support native tools
Example:
User: Calculate fibonacci sequence up to 100
Agent: I'll create and run a Python script for you.
[Creates fibonacci.py → Executes → Returns result]
Output: [1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89]
Professional cryptocurrency analyzer with real-time Binance data
Real-time data from Binance API:
- 💰 Price & Candles — Current price, 24h change, multi-timeframe data
- 📈 12 Technical Indicators — RSI, MACD, EMA, Bollinger, StochRSI, ADX, ATR, VWAP, OBV, Ichimoku, SuperTrend, CMF
- 📊 Orderbook Analysis — Deltas at 7 levels (1.5%, 3%, 5%, 15%, 30%, 60%, 90%)
- 🔄 Trade Flow — Buy/sell pressure, whale activity, trade velocity, size distribution
- 📉 Futures Data — Funding Rate, Open Interest, Long/Short Ratio
- 🎯 Smart Money Concepts — FVG, Order Blocks, Market Structure (HH/HL/LL/LH), Liquidity Zones
- 🗓️ Multi-Timeframe — Short/Medium/Long-term analysis (3 horizons)
- 📊 Volume Analysis — Volume Delta, Relative Volume, Volume Profile (POC, VAH, VAL)
- 😨 Market Context — Fear & Greed Index, total market cap
Modular Queries:
"analyze BTC" → Full analysis
"indicators ETH" → Technical indicators only
"orderbook SOL" → Orderbook analysis
"smc BTC" → Smart Money Concepts
"sentiment ETH" → Fear&Greed + Funding Rate
"volume SOL" → Volume analysis
"mtf BTC" → Multi-timeframe
Example Analysis:
📉 BTC/USDT Analysis | 2025-12-07
Price: $89,690.73 (-0.03% 24h)
MTF Signal: Bearish (Short: bullish, Medium: mixed, Long: bearish)
Order Flow: 82.6% buys — BUT whales selling (6 whale sells, 0 buys)
Smart Money: Downtrend structure (2 LH, 2 LL), 3 bearish FVG targets
Futures: L/S Ratio 2.17 (longs overloaded — squeeze risk)
Verdict: SELL — Retail buying into whale distribution
┌─────────────┐ ┌──────────────┐ ┌─────────────────┐
│ Frontend │────▶│ FastAPI │────▶│ LLM Provider │
│ (React) │ │ (routes) │ │ Gemini/OpenRouter│
└─────────────┘ └──────────────┘ └─────────────────┘
│
┌──────┴──────┐
▼ ▼
┌─────────┐ ┌─────────┐
│ Agent │ │ Session │
│ Dialog/ │ │ Manager │
│ Coder/ │ └─────────┘
│ Crypto │
└────┬────┘
▼
┌─────────┐
│ Tools │
│ search/ │
│ files/ │
│ code/ │
│ crypto │
└─────────┘
This project implements multiple layers of security for code execution:
| Layer | Protection |
|---|---|
| 🐳 Sandbox Container | Code runs in isolated Docker container with no network access |
| 📁 Path Validation | All file operations restricted to session workspace |
| ⏱️ Resource Limits | CPU, memory, and timeout constraints |
| 👤 Non-root User | Sandbox runs as unprivileged user |
⚠️ Disclaimer: This is a demonstration project. For production use, consider additional hardening (gVisor, Firecracker, etc.)
| Issue | Description | Status |
|---|---|---|
| Sandbox Fallback | If sandbox is unavailable, code may execute locally | |
| Context Window | Long conversations may exceed token limits (no RAG) | Demo scope |
| Regex JSON Fallback | Legacy fallback for non-native tool calling models | Use Gemini 2.5 Pro |
git clone https://github.com/nssanta/Multi-Agent-Router.git
cd Multi-Agent-Router
cp .env.example .env
# Add your API keys to .env
docker compose up --buildOpen: http://localhost:3000
# Required
GEMINI_API_KEY=your_gemini_key
OPENROUTER_API_KEY=your_openrouter_key
# Optional
LLM_PROVIDER=gemini # or openrouter
LLM_MODEL=gemini-2.5-pro # default model# backend/agents/my_agent/agent.py
def create_my_agent(llm_provider, session_path):
return Agent(
name="my_agent",
llm_provider=llm_provider,
instruction="Your custom prompt",
tool_definitions=[...],
)# backend/tools/my_tool.py
def my_tool(param: str) -> str:
"""Tool description for LLM"""
return f"Result: {param}"backend/
├── agents/ # Dialog, Coder, Crypto agents
│ ├── dialog/ # Conversational agent
│ ├── coder/ # Code execution agent
│ └── crypto/ # Cryptocurrency analyst
├── api/ # FastAPI routes
├── core/ # LLM providers, sessions, executor
└── tools/ # Agent tools
├── web/ # Search, page reader
└── crypto/ # Binance API, indicators
frontend/
├── src/components/ # React components
└── src/services/ # API client
Full technical documentation: TECHNICAL_DOCS.md
MIT