Verba (The Golden RAGtriever) is a community-driven open-source application by Weaviate, designed to provide an end-to-end, streamlined, and user-friendly Retrieval-Augmented Generation (RAG) experience out of the box. Users can explore datasets and extract insights locally (Ollama, HuggingFace) or through LLM providers such as Anthropic, Cohere, OpenAI, and others.
Primary use: Build customizable personal assistants around personal data to query documents, cross-reference multiple data points, or extract insights from existing knowledge bases. It combines RAG technology with Weaviate's context-aware database.
Key capabilities: Users can choose different RAG frameworks, data types, chunking and retrieval techniques, and LLM providers. It supports multiple data import methods, including importing plain text, .pdf, .csv, and other formats via Unstructured IO; importing audio files or audio from videos via AssemblyAI. It supports OpenAI models (such as Ada, GPT3, GPT4), HuggingFace local embedding models, Groq LPUs as generation engines, Novita AI, and others. Weaviate connections support local Embedded (experimental, Windows not supported), Docker, Weaviate Cloud (WCD), and custom instances (URL, PORT, API Key can be specified).
Usage: Three deployment options are provided—pip installation (requires Python >=3.10.0,<3.13.0), building from source, and Docker deployment. The Docker method can pull the imagedocker pull semitechnologies/verba, or clone the repository and rundocker build -t verba .. After startup, first select the deployment type (Local, Docker, Weaviate Cloud, or Custom), then add files, directories, or URLs through the Import Data page, and finally ask questions on the Chat page to get semantically relevant text chunks and answers generated by the selected model. The RAG pipeline can be configured in the Config tab. API keys can be set in the frontend or configured through a .env file. Specific environment variables include UNSTRUCTURED_API_KEY, UNSTRUCTURED_API_URL, ASSEMBLYAI_API_KEY, OPENAI_API_KEY, OPENAI_BASE_URL, OPENAI_EMBED_API_KEY, OPENAI_EMBED_BASE_URL, OPENAI_CUSTOM_EMBED, GROQ_API_KEY, OLLAMA_URL, and others.
Applicable scenarios: Individuals or teams that need question answering based on their own documents or knowledge bases can run it locally to ensure data privacy, or scale through cloud deployment. Multilingual support depends on the selected embedding and generation models.
Limitations: The project has stopped development and been archived, and no longer receives updates, bug fixes, security patches, or new features; issues and PRs are no longer reviewed or merged. The code is provided as-is, with no guarantee of stability, compatibility, or support. Existing deployments can continue to run, but continuous maintenance should not be expected. Weaviate Embedded is not supported on Windows and is experimental; Verba requires data to be imported in a specific format and cannot use data already in a Weaviate instance; it can only be imported through the Verba UI; when using Ollama, ensure the server is running in the background and do not mix different models (vector dimensions may differ). Images pulled from Docker Hub are vanilla versions and do not include packages such as HuggingFace, so you need to build them yourself.
