everything-ai

A collection of TypeScript applications demonstrating how to build AI-powered applications using locally hosted Large Language Models (LLMs) via Ollama.

💻 About Ollama

Strengths of Ollama

• Easy setup: Simple installation and model management
• Good model selection: Supports Llama, Mistral, CodeLlama, and many other popular models
• REST API: Makes it easy to integrate into your apps
• Resource management: Automatically handles GPU/CPU allocation
• Active community: Regular updates and good documentation

Breakdown of different Ollama models

• Llama 3.1 8B: Great balance of capability and speed, good for most agentic tasks
• Mistral 7B: Fast and capable, particularly good at following instructions
• CodeLlama: If your agents need to write/analyze code
• Llama 3.1 70B: Much more capable but requires significant hardware (32GB+ RAM)

Ollama models and hardware considerations

• 8B models: 8-16GB RAM minimum
• 13B models: 16-32GB RAM
• 70B models: 64GB+ RAM (or quantized versions with 32GB+)

To setup Ollama

• Download from the official Ollama page (Windows, MacOS, Linux)
• Once installed, you will need to install the models
• For higher end machines and setups, feel free to choose any model you like
• For lower end machines, recommend starting with the model llama3.1:8b

Note: When Ollama is set up and running, it runs a server usually at localhost:11434 that you can use

Other alternative options (if you do not want to use Ollama)

• LM Studio: Similar to Ollama but with a GUI, good for testing different models
• vLLM: Better for high-throughput production use

🎯 Project Overview

This repository contains two complementary applications that showcase different aspects of working with local LLMs:

1. Basic LLM Test App - Simple connection testing and text generation
2. Agentic AI App - Advanced agent with reasoning and tool use capabilities

🚀 Quick Start

1. Prerequisites Setup

# Install Node.js (v18+) from https://nodejs.org
# Install Ollama from https://ollama.com

# Download and start a model
ollama pull llama3.1:8b
ollama serve

2. Project Setup

# Clone or download this project
npm install

3. Run Applications

# Test basic LLM connection
npm run dev

# Run the agentic AI chat
npm run dev:agent

📱 Applications

🧪 Basic LLM Test App

Purpose: Test your Ollama setup and understand basic LLM interaction

Features:

• ✅ Connection testing
• 📋 Model discovery
• 💬 Simple text generation
• 🔧 Error handling

Best for:

• First-time Ollama users
• Testing new models
• Basic integration examples
• Learning LLM API basics

---

🤖 Agentic AI App

Purpose: Demonstrate advanced AI agent capabilities with reasoning and tool use

Features:

• 🧠 Multi-step reasoning (ReAct pattern)
• 🛠️ Tool integration (calculator, time, weather)
• 🔄 Action chaining
• 💭 Transparent thinking process
• 💬 Interactive chat interface

Best for:

• Learning agentic AI concepts
• Building complex AI applications
• Understanding ReAct pattern
• Tool-augmented AI systems

🎓 Learning Path

Beginner Path

1. Start with Basic Test - Understand LLM basics
2. Read the code - Learn TypeScript + Ollama integration
3. Experiment with models - Try different LLMs
4. Modify prompts - See how responses change

Intermediate Path

1. Run Agentic AI - Experience tool-augmented AI
2. Study ReAct pattern - Understand reasoning + acting
3. Add custom tools - Extend agent capabilities
4. Experiment with workflows - Chain multiple actions

Advanced Path

1. Integrate real APIs - Replace mock tools
2. Add memory/state - Build stateful agents
3. Create web interfaces - Build full applications
4. Deploy at scale - Production considerations

🎯 Use Cases

Basic App Use Cases

• ✅ Testing new Ollama installations
• 🔍 Exploring different LLM models
• 📚 Learning LLM API integration
• 🛠️ Building simple AI-powered features

Agentic AI Use Cases

• 🤖 Virtual assistants with tool access
• 📊 Data analysis workflows
• 🔄 Multi-step automation
• 🧠 Complex reasoning applications
• 🛠️ Tool-augmented problem solving