Verba — The Golden RAGtriever by Weaviate

What it is

Verba is an open-source RAG (Retrieval-Augmented Generation) chatbot from the Weaviate team. It bundles document ingestion, chunking, embedding, vector search, and chat into a single application. Upload PDFs, DOCX, Markdown, plain text, web URLs, or GitHub repos, and Verba handles the entire pipeline from ingestion to conversational retrieval.

Verba is designed for teams who want to evaluate RAG patterns, demo to stakeholders, or self-host a private knowledge assistant without writing custom code.

How it saves time or tokens

Building a RAG application from scratch requires choosing and integrating a vector store, embedding model, chunking strategy, and chat UI. Verba provides all of these in one package. The zero-infra mode runs Weaviate Embedded locally, so you do not need a separate database server. Pluggable components let you swap models and embedders without changing application code.

How to use

1. Install and run Verba:

pip install goldenverba
verba start
# Open http://localhost:8000

1. Or run via Docker Compose:

git clone https://github.com/weaviate/Verba
cd Verba
docker compose up -d

1. Upload documents through the web UI and start chatting with your knowledge base.

Example

Configuring Verba to use Ollama for local, private RAG:

# Set environment variables before starting
export OLLAMA_URL='http://localhost:11434'
export OLLAMA_MODEL='llama3'
export OLLAMA_EMBED_MODEL='nomic-embed-text'

# Start Verba with Ollama as the backend
verba start

The architecture follows a pluggable pipeline:

[Document Upload] -> [Reader] -> [Chunker] -> [Embedder] -> [Weaviate Vector Store]
                                                                      |
[Chat UI] <- [LLM Generator] <- [Semantic Search] <------------------+

Related on TokRepo

• RAG tools — retrieval-augmented generation frameworks and tools
• Self-hosted tools — privacy-focused local deployments

Common pitfalls

• Large PDF uploads without chunking configuration can exceed context windows; adjust chunk size in the settings for documents over 100 pages
• The default Weaviate Embedded mode stores data in memory; restart loses all data unless you configure persistent storage
• OpenAI API keys must be set as environment variables before starting; Verba does not prompt for them at runtime