Dynamic Tool Attachment Project

A TypeScript-based project demonstrating dynamic tool attachment using in-memory RAG (Retrieval-Augmented Generation). This project shows how to dynamically select and attach relevant tools to an LLM based on user input using semantic search.

Uses Ollama as the LLM provider with llama3.2 3b. This is a companion project to the Dynamic Tool Attachment blog post.

How it Works

The project embeds tool descriptions using the nomic-embed-text model and matches user queries to the most relevant tools using cosine similarity. This allows for dynamic, context-aware tool selection during conversations.

Prerequisites

• Node.js and npm installed
• Ollama running locally with the llama3.2 (3b) and nomic-embed-text models

Setup

1. Install dependencies:

npm install

2. Make sure Ollama is running with the required models

Usage

Run the project:

tsx index.ts

Enter messages in the console to interact with the LLM. The system will automatically attach the most relevant tools based on your input.

Enter your message:
find hammer with ID 123
Attaching tool: { tool: 'findTool', similarity: 0.5955042990419085 }
Attaching tool: { tool: 'findToy', similarity: 0.5411030864968827 }
Attaching tool: { tool: 'findMovie', similarity: 0.5219368182792513 }
Attaching tool: { tool: 'findSong', similarity: 0.5155495695032712 }
Attaching tool: { tool: 'findCar', similarity: 0.5132604044773412 }
Tools called:  [ { function: { name: 'findTool', arguments: [Object] } } ]

License

This project is licensed under the MIT License.