Verba: A Discontinued RAG Toolkit That’s Actually Worth Forking
Sometimes the best tools are the ones that had a short shelf life. Verba is exactly that. It was a fully open source RAG (Retrieval Augmented Generation) toolkit built by the Weaviate team. And while it’s officially discontinued, the code is still live on GitHub — and it’s surprisingly complete.
If you’ve ever tried to build a RAG pipeline from scratch, you know it’s not trivial. You need chunking, embedding, retrieval, generation, and some kind of UI to test it. Verba gives you all of that in a single package. It’s a great reference implementation, or even a foundation to fork and customize.
What It Does
Verba is a turnkey RAG application. You feed it documents (PDF, text, markdown, etc.), it chunks and embeds them into Weaviate (their vector database), and then you can query the documents with a conversational interface. The whole thing runs locally or on your own infra.
Under the hood, it handles:
- Document ingestion and chunking
- Embedding generation (supports OpenAI, Cohere, and local models)
- Hybrid search (vector + keyword)
- A simple Svelte frontend for chatting with your documents
There’s no magic. It’s just a clean, opinionated pipeline that works out of the box.
Why It’s Cool
1. Full stack, minimal config
You don’t need to glue together LangChain, a vector DB, a frontend, and a search API. Verba is one repo with a clear docker-compose.yml. Start it up, upload a PDF, and you’re chatting with it in minutes.
2. Supports multiple LLM providers
It works with OpenAI, Cohere, and even local models via Ollama. That’s rare for a beginner-friendly toolkit. You can swap your backend without changing the application code.
3. Good for learning and reference
The codebase is well structured. If you want to understand how a real RAG system handles chunking, metadata extraction, or hybrid search, Verba is a clean example. It’s not overengineered.
4. Great for prototyping
If you need to quickly test a RAG pipeline for a specific document set or use case, Verba is faster than cobbling things together yourself. Just fork it, tweak the chunk size, and go.
5. It’s abandoned, not broken
The repository still works. The dependencies are pinned. The docs are intact. It’s not rotting — it’s just not getting new features. That’s fine for a foundation to build on.
How to Try It
-
Clone the repo:
git clone https://github.com/weaviate/Verba -
Follow the quickstart in the README (it uses Docker Compose).
docker compose up -
Open the UI at
http://localhost:8080. -
Upload a document and start asking questions.
No account required if you use the local LLM option. You can also plug in an OpenAI key if you want to test with GPT-4.
Final Thoughts
I’d normally hesitate to recommend a discontinued project, but Verba is an exception. It’s not a half-baked prototype — it’s a polished, tested RAG toolkit that the Weaviate team clearly put real effort into. The official support is gone, but the code works today and will work for a long time given it’s sandboxed in its dependencies.
If you’re building a RAG system for a specific use case (internal docs, research papers, legal documents, etc.), Verba is a fantastic starting point. Fork it, rip out the parts you don’t need, or just use it to learn how real RAG pipelines are wired together.
It’s rare to get a complete, production-grade open source project for free. Grab it while it’s still warm.
Found this project on GitHub? Follow @githubprojects for more developer tools and repos worth exploring.