Chat Agent - AI Assistant with Tools

A modern web-based chat application that allows you to interact with AI agents powered by open-source LLMs through Ollama. The agents have access to tools and can perform actions like web browsing.

Features

🤖 AI Agent Integration: Powered by Ollama with configurable open-source models
🛠️ Native Tool Support: Uses Ollama's built-in tool calling API for reliable function execution
💬 Modern Chat UI: Clean, responsive interface with real-time streaming
📚 Session Management: Local storage of chat sessions with full history
🎨 Markdown Support: Rich text rendering with syntax highlighting
🌐 Tool Visualization: Interactive displays for tool executions and results
⚡ Streaming Tools: Real-time tool execution with streaming responses
🐳 Docker Ready: Full containerization with Docker Compose
🚀 Production Ready: Nginx-served frontend with Traefik labels

Quick Start

Prerequisites

Docker and Docker Compose
At least 4GB RAM for model execution

⚡ New in v2.0: Native Ollama tool calling support! The agent now uses Ollama's built-in function calling API for more reliable and faster tool execution with streaming support.

1. Clone and Setup

git clone <your-repo>
cd chat-agent

# Option 1: Create .env file (recommended for development)
cp .env.example .env
# Edit .env with your configuration

# Option 2: Use environment variables directly (recommended for production)
export OLLAMA_URL=http://ollama:11434
# OR
export ZHIPUAI_API_KEY=your_zhipu_api_key

2. Start Services

docker-compose up -d

This will start:

Ollama on port 11434
Backend API on port 8000
Frontend on port 3000

3. Pull a Model

# Pull a model (this may take a few minutes)
docker exec ollama ollama pull llama3.2

# Or try other models with tool support:
# docker exec ollama ollama pull llama3.1
# docker exec ollama ollama pull qwen2.5

4. Access the Application

Open your browser and go to http://localhost:3000

Configuration

Environment Variables

You can configure the application using environment variables directly or with a .env file. At minimum, you need one of the following LLM providers:

Required (at least one):

# For local Ollama instance
OLLAMA_URL=http://ollama:11434

# OR for Zhipu AI cloud service
ZHIPUAI_API_KEY=your_zhipu_api_key

Optional configuration:

Edit .env to configure additional options:

# LLM Model Configuration
LLM_MODEL=llama3.2                  # Choose your model
OLLAMA_URL=http://ollama:11434      # Ollama service URL
CORS_ORIGINS=http://localhost:3000,https://ccoreilly.github.io  # Allowed origins

# Agent Configuration
AGENT_TYPE=softcatala_english       # Agent type: softcatala_english (default) or softcatala_catalan

# Production (for Traefik)
# TRAEFIK_HOST=your-domain.com

Agent Types

The system supports two different Softcatalà agent types that can be configured using the AGENT_TYPE environment variable:

🏴‍☠️ softcatala_english (Default)

System Prompt: English (for better LLM performance)
Response Language: Catalan only
Purpose: Optimized Softcatalà assistant with English instructions for better model comprehension
Best For: Production use with Catalan language support

🏴‍☠️ softcatala_catalan

System Prompt: Catalan
Response Language: Catalan only
Purpose: Fully Catalan Softcatalà assistant
Best For: Testing prompt effectiveness in Catalan vs English

Usage Examples:

# Use default Softcatalà agent (English prompt, Catalan responses)
AGENT_TYPE=softcatala_english

# Use fully Catalan Softcatalà agent
AGENT_TYPE=softcatala_catalan

Available Models

Popular models you can use:

llama3.1 - Latest model with native tool support (4.7GB) ⭐
llama3.2 - Efficient with tool calling (2.0GB) ⭐
qwen2.5 - Advanced model with excellent tool support (4.4GB) ⭐
mistral - Fast and efficient (4.1GB)
codellama - Code-focused (3.8GB)
llama2:13b - Larger model for better quality (7.3GB)

⭐ Recommended for tool usage - These models have native training for function calling and provide the best experience with tools.

Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Frontend      │    │    Backend      │    │     Ollama      │
│   (React)       │◄──►│   (FastAPI)     │◄──►│   (LLM Server)  │
│   Port 3000     │    │   Port 8000     │    │   Port 11434    │
└─────────────────┘    └─────────────────┘    └─────────────────┘

Components

Frontend: React TypeScript app with local storage for chat sessions
Backend: FastAPI server with streaming responses and tool execution
Agent: LLM integration with tool calling capabilities
Tools: Extensible system with web browsing tool included
Ollama: Local LLM server for model inference

API Documentation

Endpoints

GET / - API status
GET /health - Health check with model info
POST /chat/stream - Streaming chat endpoint
GET /models - List available models

Chat API Example

curl -X POST http://localhost:8000/chat/stream \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "Hello!", "timestamp": "2023-01-01T00:00:00Z"}
    ],
    "session_id": "session-123"
  }'

Tool System

Available Tools

Web Browser Tool

Name: web_browser
Description: Browse web pages and extract content
Parameters:
- url (required): URL to browse
- extract_links (optional): Extract links from page
- max_content_length (optional): Limit content length

Note: Tools are now integrated using Ollama's native tool calling API instead of prompt-based approaches, providing better reliability and streaming support.

Adding Custom Tools

Create a new tool class in backend/tools/:

from .base import BaseTool, ToolDefinition, ToolParameter

class MyCustomTool(BaseTool):
    @property
    def definition(self) -> ToolDefinition:
        return ToolDefinition(
            name="my_tool",
            description="Description of what the tool does",
            parameters=[
                ToolParameter(
                    name="param1",
                    type="string", 
                    description="Parameter description",
                    required=True
                )
            ]
        )
    
    async def execute(self, **kwargs):
        # Tool implementation
        return {"result": "success"}

from tools.my_custom_tool import MyCustomTool

my_tool = MyCustomTool()
agent = Agent(tools=[web_browser_tool, my_tool])

Development

Local Development

For development with hot reload:

# Backend
cd backend
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

# Frontend  
cd frontend
npm install
npm start

Project Structure

chat-agent/
├── backend/                 # FastAPI backend
│   ├── main.py             # Main application
│   ├── agent.py            # Agent implementation
│   ├── tools/              # Tool implementations
│   │   ├── base.py         # Base tool class
│   │   └── web_browser.py  # Web browsing tool
│   ├── requirements.txt    # Python dependencies
│   └── Dockerfile         # Backend container
├── frontend/               # React frontend
│   ├── src/
│   │   ├── components/     # UI components
│   │   ├── hooks/         # Custom hooks
│   │   ├── utils/         # Utilities
│   │   └── types.ts       # TypeScript types
│   ├── package.json       # Node dependencies
│   └── Dockerfile        # Frontend container
├── docker-compose.yml     # Service orchestration
└── .env.example          # Environment template

Production Deployment

With Traefik

The services include Traefik labels for production deployment:

# docker-compose.prod.yml
version: '3.8'
services:
  frontend:
    labels:
      - "traefik.enable=true"
      - "traefik.http.routers.chat-frontend.rule=Host(`your-domain.com`)"
      - "traefik.http.routers.chat-frontend.entrypoints=websecure"
      - "traefik.http.routers.chat-frontend.tls.certresolver=letsencrypt"

Environment Variables for Production

LLM_MODEL=llama3.2
OLLAMA_URL=http://ollama:11434
CORS_ORIGINS=https://your-domain.com
TRAEFIK_HOST=your-domain.com
AGENT_TYPE=softcatala_english

Troubleshooting

Common Issues

Ollama not accessible

docker exec ollama ollama list  # Check if models are loaded
docker logs ollama              # Check Ollama logs

Frontend can't connect to backend
- Check CORS_ORIGINS in environment
- Verify backend is running on port 8000

Model not found

docker exec ollama ollama pull llama3.2  # Pull the model

Out of memory
- Try smaller models like mistral
- Increase Docker memory limit

Performance Tips

Use faster models like mistral for better response times
Increase model context length for longer conversations
Use SSD storage for faster model loading

Contributing

Fork the repository
Create a feature branch
Add your tool or feature
Submit a pull request

License

MIT License - see LICENSE file for details

ccoreilly/agent-softcatala