rfaiss

rfaiss is an experimental R wrapper around FAISS, Facebook AI’s library for efficient similarity search and clustering of dense vectors.
It is designed for users who are already familiar with Python embeddings workflows (e.g., sentence-transformers) and want to experiment with vector search inside R via reticulate.

⚠️ Important:

• This package is experimental and not intended for general use.
• It is not on CRAN and likely never will be.
• Expect rough edges, higher memory usage, and a dependency on a functioning Python + FAISS installation.

---

Why use rfaiss?

• Direct access to a true vector store: Unlike some R-native approaches, FAISS is a production-grade library widely used in large-scale AI applications.
• Reticulate bridge: Lets you tap into the FAISS Python ecosystem directly from R, without having to leave your R workflow.
• Good for experimentation: Useful for prototyping retrieval-augmented generation (RAG) workflows in R when you already rely on embeddings generated with Python models.

---

Why not use rfaiss?

• Memory usage: FAISS indexes are fast but RAM-hungry. Even small experiments can use a decent amount of memory.
• Experimental state: API stability is not guaranteed.
• Alternative in R: The tidyverse team is building ragnar, which uses DuckDB for embedding workflows. This will likely be the long-term ecosystem standard for R users.
• CRAN status: rfaiss will probably never be on CRAN, given its reliance on Python and FAISS.

---

Installation

Because rfaiss depends on Python, FAISS, and reticulate, you must ensure these are available in your environment.

# Install from GitHub
remotes::install_github("lazasaurus-ai//rfaiss")

# Load the package
library(rfaiss)

On the Python side you need:

pip install faiss-cpu sentence-transformers

Example

library(rfaiss)

# Create some random embeddings
set.seed(123)
mat <- matrix(rnorm(100 * 128), nrow = 100, ncol = 128)

# Build an index
index <- faiss_index(mat)

# Query the index
query <- matrix(rnorm(1 * 128), nrow = 1)
faiss_search(index, query, k = 5)

Notes

• Ensure your Python environment is correctly configured with reticulate::use_virtualenv() or reticulate::use_condaenv().

• Best used by those who already work with sentence-transformers in Python and simply want to experiment with FAISS from R.

• For R-native workflows, keep an eye on ragnar.