wizrd: An R Interface to Large Language Models

Overview

The wizrd package enables embedding Large Language Models (LLMs) in R programs by treating them as functions with structured inputs and outputs. It supports local and remote LLMs, including Ollama, llama.cpp, OpenAI, and Azure, via an extensible abstraction. Features include prompt templating, tool calling, and experimental support for text embedding, vector stores, and Results Augmented Generation (RAG).

Installation

install.packages("remotes")
remotes::install_github("lawremi/wizrd")

Quick Start

The following example runs out of the box using Mozilla's llamafile framework (no system dependencies):

library(wizrd)

# Use a local Llama 3.2 3B model via llamafile (auto-downloads if needed)
agent <- llamafile_llama()
predict(agent, "Describe the mtcars dataset")

For advanced usage, including parameterized prompts, tool calling, structured output, and retrieval-augmented generation (RAG), see the package vignette:

vignette("wizrd")

Documentation and Support

• Reference: help(package = "wizrd")
• Website: https://lawremi.github.io/wizrd/
• GitHub Issues: https://github.com/lawremi/wizrd/issues