anikchand461/ragbucket

★ 0HTMLAudience · developerComplexity · 2/5ActiveLicenseSetup · easy

Mindmap

mindmap
  root((ragbucket))
    Inputs
      Folder of documents
      Embedding provider config
      Chunk size and overlap
    Outputs
      Single .rag file
      FAISS index
      Chunked JSON
      Answers via ask
    Use Cases
      Ship a RAG corpus as one file
      Swap embeddings and generation
      Local on-device RAG
    Tech Stack
      Python
      FAISS
      SentenceTransformers
      Groq
      OpenAI

mindmap root((ragbucket)) Inputs Folder of documents Embedding provider config Chunk size and overlap Outputs Single .rag file FAISS index Chunked JSON Answers via ask Use Cases Ship a RAG corpus as one file Swap embeddings and generation Local on-device RAG Tech Stack Python FAISS SentenceTransformers Groq OpenAI

Things people build with this

USE CASE 1

Build a portable .rag file from a folder of documents and ship it to teammates

USE CASE 2

Swap between Groq, OpenAI, Gemini, or Anthropic for generation without rebuilding the index

USE CASE 3

Run a fully local RAG pipeline with Sentence Transformers embeddings

USE CASE 4

Embed a prebuilt knowledge base inside a desktop or CLI tool with one file

Tech stack

PythonFAISSSentenceTransformersGroqOpenAI

Getting it running

Difficulty · easy Time to first run · 30min

Base install is light, but using local embeddings pulls in heavy Sentence Transformers dependencies you opt into explicitly.

MIT license, so you can use, modify, and redistribute the code freely as long as you keep the copyright notice.

In plain English

RagBucket is a Python library that tries to make Retrieval-Augmented Generation systems portable. A RAG system is the common setup where you take a pile of documents, break them into chunks, turn the chunks into numerical embeddings, and store those in a vector index so a language model can later look up relevant pieces before answering a question. The author's complaint is that every machine learning model format, such as .pt.onnx.gguf, and .h5, is portable by default, but a typical RAG pipeline is spread across vector databases, embedding scripts, chunking configs, and provider integrations that all have to be rebuilt when you move between machines. The library's answer is a single file format called .rag. A .rag artifact bundles three things into one file: a FAISS vector index, the chunked documents as JSON, and a manifest that records the embedding configuration, model info, and version. The pitch is that you build it once, ship the file around, and then load and query it anywhere with one line of code, with no external vector database to set up. The quickstart shows two short Python scripts. The first uses a RagBuilder with a RagConfig that picks an embedding provider, chunk size and overlap, and top_k retrieval setting, and then writes the .rag file from a folder of documents. The second uses a RagRuntime that loads the .rag file, attaches a generation provider such as Groq with a Llama 3.1 model, and exposes a rag.ask method that takes a question and returns an answer. RagBucket cleanly separates retrieval from generation, so the embedding and generation sides can be mixed and matched. Supported generation providers include Groq, OpenAI, Gemini, and Anthropic, with example models listed for each. Supported embedding providers include a local Sentence Transformers option, Cohere, OpenAI, Gemini, and Voyage. The base install stays light: heavy local embedding dependencies are only pulled in if you set the embedding provider to local. Installation is through uv or pip as the ragbucket package, the project is MIT licensed, and the repository has zero stars at the time of writing.

Copy-paste prompts

Prompt 1

Use ragbucket to build a .rag artifact from my docs folder with SentenceTransformers and a 512 chunk size

Prompt 2

Load an existing .rag file at runtime, attach Groq with Llama 3.1, and answer a question via rag.ask

Prompt 3

Compare base install size when I keep the embedding provider remote versus switching to local

Prompt 4

Show how to inspect the manifest inside a .rag file to verify embedding model and version

Prompt 5

Wire ragbucket into a FastAPI endpoint so a frontend can query a shipped .rag knowledge base

Open on GitHub → Explain another repo

Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.