Welcome to Verba: The Golden RAGtriever, an community-driven open-source application designed to offer an end-to-end, streamlined, and user-friendly interface for Retrieval-Augmented Generation (RAG) out of the box. In just a few easy steps, explore your datasets and extract insights with ease, either locally with Ollama and Huggingface or through LLM providers such as Anthrophic, Cohere, and OpenAI. This project is built with and for the community, please be aware that it might not be maintained with the same urgency as other Weaviate production applications. Feel free to contribute to the project and help us make Verba even better! <3
pip install goldenverba
- Verba
- ✨ Getting Started with Verba
- 🔑 API Keys
- Quickstart: Deploy with pip
- Quickstart: Build from Source
- Quickstart: Deploy with Docker
- 💾 Verba Walkthrough
- 💖 Open Source Contribution
- 🚩 Known Issues
- ❔FAQ< 8000 /li>
Verba is a fully-customizable personal assistant utilizing Retrieval Augmented Generation (RAG) for querying and interacting with your data, either locally or deployed via cloud. Resolve questions around your documents, cross-reference multiple data points or gain insights from existing knowledge bases. Verba combines state-of-the-art RAG techniques with Weaviate's context-aware database. Choose between different RAG frameworks, data types, chunking & retrieving techniques, and LLM providers based on your individual use-case.
Weaviate is proud to offer this open-source project for the community. While we strive to address issues as fast as we can, please understand that it may not be maintained with the same rigor as production software. We welcome and encourage community contributions to help keep it running smoothly. Your support in fixing open issues quickly is greatly appreciated.
| 🤖 Model Support | Implemented | Description |
|---|---|---|
| Ollama (e.g. Llama3) | ✅ | Local Embedding and Generation Models powered by Ollama |
| HuggingFace (e.g. MiniLMEmbedder) | ✅ | Local Embedding Models powered by HuggingFace |
| Cohere (e.g. Command R+) | ✅ | Embedding and Generation Models by Cohere |
| Anthrophic (e.g. Claude Sonnet) | ✅ | Embedding and Generation Models by Anthrophic |
| OpenAI (e.g. GPT4) | ✅ | Embedding and Generation Models by OpenAI |
| Groq (e.g. Llama3) | ✅ | Generation Models by Groq (LPU inference) |
| Upstage (e.g. Solar) | ✅ | Embedding and Generation Models by Upstage |
| 🤖 Embedding Support | Implemented | Description |
|---|---|---|
| Weaviate | ✅ | Embedding Models powered by Weaviate |
| Ollama | ✅ | Local Embedding Models powered by Ollama |
| SentenceTransformers | ✅ | Embedding Models powered by HuggingFace |
| Cohere | ✅ | Embedding Models by Cohere |
| VoyageAI | ✅ | Embedding Models by VoyageAI |
| OpenAI | ✅ | Embedding Models by OpenAI |
| Upstage | ✅ | Embedding Models by Upstage |
| 📁 Data Support | Implemented | Description |
|---|---|---|
| UnstructuredIO | ✅ | Import Data through Unstructured |
| Firecrawl | ✅ | Scrape and Crawl URL through Firecrawl |
| UpstageDocumentParse | ✅ | Parse Documents through Upstage Document AI |
| PDF Ingestion | ✅ | Import PDF into Verba |
| GitHub & GitLab | ✅ | Import Files from Github and GitLab |
| CSV/XLSX Ingestion | ✅ | Import Table Data into Verba |
| .DOCX | ✅ | Import .docx files |
| Multi-Modal (using AssemblyAI) | ✅ | Import and Transcribe Audio through AssemblyAI |
| ✨ RAG Features | Implemented | Description |
|---|---|---|
| Hybrid Search | ✅ | Semantic Search combined with Keyword Search |
| Autocomplete Suggestion | ✅ | Verba suggests autocompletion |
| Filtering | ✅ | Apply Filters (e.g. documents, document types etc.) before performing RAG |
| Customizable Metadata | ✅ | Free control over Metadata |
| Async Ingestion | ✅ | Ingest data asynchronously to speed up the process |
| Advanced Querying | planned ⏱️ | Task Delegation Based on LLM Evaluation |
| Reranking | planned ⏱️ | Rerank results based on context for improved results |
| RAG Evaluation | planned ⏱️ | Interface for Evaluating RAG pipelines |
| Agentic RAG | out of scope ❌ | Agentic RAG pipelines |
| Graph RAG | out of scope ❌ | Graph-based RAG pipelines |
| 🗡️ Chunking Techniques | Implemented | Description |
|---|---|---|
| Token | ✅ | Chunk by Token powered by spaCy |
| Sentence | ✅ | Chunk by Sentence powered by spaCy |
| Semantic | ✅ | Chunk and group by semantic sentence similarity |
| Recursive | ✅ | Recursively chunk data based on rules |
| HTML | ✅ | Chunk HTML files |
| Markdown | ✅ | Chunk Markdown files |
| Code | ✅ | Chunk Code files |
| JSON | ✅ | Chunk JSON files |
| 🆒 Cool Bonus | Implemented | Description |
|---|---|---|
| Docker Support | ✅ | Verba is deployable via Docker |
| Customizable Frontend | ✅ | Verba's frontend is fully-customizable via the frontend |
| Vector Viewer | ✅ | Visualize your data in 3D |
| Multi-User Collaboration | out of scope ❌ | Multi-User Collaboration in Verba |
| 🤝 RAG Libraries | Implemented | Description |
|---|---|---|
| LangChain | ✅ | Implement LangChain RAG pipelines |
| Haystack | planned ⏱️ | Implement Haystack RAG pipelines |
| LlamaIndex | planned ⏱️ | Implement LlamaIndex RAG pipelines |
Something is missing? Feel free to create a new issue or discussion with your idea!
You have three deployment options for Verba:
- Install via pip
pip install goldenverba
- Build from Source
git clone https://github.com/weaviate/Verba
pip install -e .
- Use Docker for Deployment
Prerequisites: If you're not using Docker, ensure that you have Python >=3.10.0,<3.13.0 installed on your system.
git clone https://github.com/weaviate/Verba
docker compose --env-file <your-env-file> up -d --build
If you're unfamiliar with Python and Virtual Environments, please read the python tutorial guidelines.
You can set all API keys in the Verba frontend, but to make your life easier, we can also prepare a .env file in which Verba will automatically look for the keys. Create a .env in the same directory you want to start Verba in. You can find an .env.example file in the goldenverba directory.
Make sure to only set environment variables you intend to use, environment variables with missing or incorrect values may lead to errors.
Below is a comprehensive list of the API keys and variables you may require:
| Environment Variable | Value | Description |
|---|---|---|
| WEAVIATE_URL_VERBA | URL to your hosted Weaviate Cluster | Connect to your WCS Cluster |
| WEAVIATE_API_KEY_VERBA | API Credentials to your hosted Weaviate Cluster | Connect to your WCS Cluster |
| ANTHROPIC_API_KEY | Your Anthropic API Key | Get Access to Anthropic Models |
| OPENAI_API_KEY | Your OpenAI Key | Get Access to OpenAI Models |
| OPENAI_BASE_URL | URL to OpenAI instance | Models |
| COHERE_API_KEY | Your API Key | Get Access to Cohere Models |
| GROQ_API_KEY | Your Groq API Key | Get Access to Groq Models |
| OLLAMA_URL | URL to your Ollama instance (e.g. http://localhost:11434 ) | Get Access to Ollama Models |
| UNSTRUCTURED_API_KEY | Your API Key | Get Access to Unstructured Data Ingestion |
| UNSTRUCTURED_API_URL | URL to Unstructured Instance | Get Access to Unstructured Data Ingestion |
| ASSEMBLYAI_API_KEY | Your API Key | Get Access to AssemblyAI Data Ingestion |
| GITHUB_TOKEN | Your GitHub Token | Get Access to Data Ingestion via GitHub |
| GITLAB_TOKEN | Your GitLab Token | Get Access to Data Ingestion via GitLab |
| FIRECRAWL_API_KEY | Your Firecrawl API Key | Get Access to Data Ingestion via Firecrawl |
| VOYAGE_API_KEY | Your VoyageAI API Key | Get Access to Embedding Models via VoyageAI |
| EMBEDDING_SERVICE_URL | URL to your Embedding Service Instance | Get Access to Embedding Models via Weaviate Embedding Service |
| EMBEDDING_SERVICE_KEY | Your Embedding Service Key | Get Access to Embedding Models via Weaviate Embedding Service |
| UPSTAGE_API_KEY | Your Upstage API Key | Get Access to Upstage Models |
| UPSTAGE_BASE_URL | URL to Upstage instance | Models |
| DEFAULT_DEPLOYMENT | Local, Weaviate, Custom, Docker | Set the default deployment mode |
Verba provides flexibility in connecting to Weaviate instances based on your needs. You have three options:
- Local Deployment: Use Weaviate Embedded which runs locally on your device (except Windows, choose the Docker/Cloud Deployment)
- Docker Deployment: Choose this option when you're running Verba's Dockerfile.
- Cloud Deployment: Use an existing Weaviate instance hosted on WCD to run Verba
💻 Weaviate Embedded
Embedded Weaviate is a deployment model that runs a Weaviate instance from your application code rather than from a stand-alone Weaviate server installation. When you run Verba in Local Deployment, it will setup and manage Embedded Weaviate in the background. Please note that Weaviate Embedded is not supported on Windows and is in Experimental Mode which can bring unexpected errors. We recommend using the Docker Deployment or Cloud Deployment instead. You can read more about Weaviate Embedded here.
🌩️ Weaviate Cloud Deployment (WCD)
If you prefer a cloud-based solution, Weaviate Cloud (WCD) offers a scalable, managed environment. Learn how to set up a cloud cluster and get the API keys by following the Weaviate Cluster Setup Guide.
🐳 Docker Deployment Another local alternative is deploying Weaviate using Docker. For more details, follow the How to install Verba with Docker section.
⚙️ Custom Weaviate Deployment
If you're hosting Weaviate yourself, you can use the Custom deployment option in Verba. This will allow you to specify the URL, PORT, and API key of your Weaviate instance.
Verba supports Ollama models. Download and Install Ollama on your device (https://ollama.com/download). Make sure to install your preferred LLM using ollama run <model>.
Tested with llama3, llama3:70b and mistral. The bigger models generally perform better, but need more computational power.
Make sure Ollama Server runs in the background and that you don't ingest documents with different ollama models since their vector dimension can vary that will lead to errors
You can verify that by running the following command
ollama run llama3
Verba supports importing documents through Unstructured IO (e.g plain text, .pdf, .csv, and more). To use them you need the UNSTRUCTURED_API_KEY and UNSTRUCTURED_API_URL environment variable. You can get it from Unstructured
UNSTRUCTURED_API_URL is set to
https://api.unstructuredapp.io/general/v0/generalby default
Verba supports importing documents through AssemblyAI (audio files or audio from video files). To use them you need the ASSEMBLYAI_API_KEY environment variable. You can get it from AssemblyAI
Verba supports OpenAI Models such as Ada, GPT3, and GPT4. To use them, you need to specify the OPENAI_API_KEY environment variable. You can get it from OpenAI
You can also add a OPENAI_BASE_URL to use proxies such as LiteLLM (https://github.com/BerriAI/litellm)
OPENAI_BASE_URL=YOUR-OPENAI_BASE_URL
If you want to use the HuggingFace Features, make sure to install the correct Verba package. It will install required packages to use the local embedding models. Please note that on startup, Verba will automatically download and install embedding models when used.
pip install goldenverba[huggingface]
or
pip install `.[huggingface]`If you're using Docker, modify the
Dockerfileaccordingly. It's not possible to install a custom Verba installation if you pull the Docker Image from the Docker Hub, as of now, you'd need to install the Docker deployment from the source code and modify theDockerfilebeforehand.
To use Groq LPUs as generation engine, you need to get an API key from Groq.
Although you can provide it in the graphical interface when Verba is up, it is recommended to specify it as
GROQ_API_KEYenvironment variable before you launch the application.
It will allow you to choose the generation model in an up-to-date available models list.
Python >=3.10.0
- (Very Important) Initialize a new Python Environment
python3 -m virtualenv venv
source ve
8000
nv/bin/activate
- Install Verba
pip install goldenverba
- Launch Verba
verba start
You can specify the --port and --host via flags
- Access Verba
Visit localhost:8000
- (Optional)Create .env file and add environment variables
- Clone the Verba repos
git clone https://github.com/weaviate/Verba.git
- Initialize a new Python Environment
python3 -m virtualenv venv
source venv/bin/activate
- Install Verba
pip install -e .
- Launch Verba
verba start
You can specify the --port and --host via flags
- Access Verba
Visit localhost:8000
- (Optional) Create .env file and add environment variables
Docker is a set of platform-as-a-service products that use OS-level virtualization to deliver software in packages called containers. To get started with deploying Verba using Docker, follow the steps below. If you need more detailed instructions on Docker usage, check out the Docker Curriculum.
You can use docker pull semitechnologies/verba to pull the latest Verba Docker Image. Please note, that by pulling directly from Docker Hub you're only able to install the vanilla Verba version that does not include packages e.g HuggingFace. If you want to use Docker and HuggingFace please follow the steps below.
To build the image yourself, you can clone the Verba repository and run docker build -t verba . inside the Verba directory.
- Clone the Verba repos Ensure you have Git installed on your system. Then, open a terminal or command prompt and run the following command to clone the Verba repository:
git clone https://github.com/weaviate/Verba.git
-
Set necessary environment variables Make sure to set your required environment variables in the
.envfile. You can read more about how to set them up in the API Keys Section -
Adjust the docker-compose file You can use the
docker-compose.ymlto add required environment variables under theverbaservice and can also adjust the Weaviate Docker settings to enable Authentification or change other settings of your database instance. You can read more about the Weaviate configuration in our docker-compose documentation. You can also uncomment theollamaservice to use Ollama within the same docker compose.
Please make sure to only add environment variables that you really need.
- Deploy using Docker With Docker installed and the Verba repository cloned, navigate to the directory containing the Docker Compose file in your terminal or command prompt. Run the following command to start the Verba application in detached mode, which allows it to run in the background:
docker compose up -d
docker compose --env-file goldenverba/.env up -d --build
This command will download the necessary Docker images, create containers, and start Verba. Remember, Docker must be installed on your system to use this method. For installation instructions and more details about Docker, visit the official Docker documentation.
- Access Verba
-
You can access your local Weaviate instance at
localhost:8080 -
You can access the Verba frontend at
localhost:8000
If you want your Docker Instance to install a specific version of Verba you can edit the Dockerfile and change the installation line.
RUN pip install -e '.'
The first screen you'll see is the deployment screen. Here you can select between Local, Docker, Weaviate Cloud, or Custom deployment. The Local deployment is using Weaviate Embedded under the hood, which initializes a Weaviate instance behind the scenes. The Docker deployment is using a separate Weaviate instance that is running inside the same Docker network. The Weaviate Cloud deployment is using a Weaviate instance that is hosted on Weaviate Cloud Services (WCS). The Custom deployment allows you to specify your own Weaviate instance URL, PORT, and API key.
You can skip this part by setting the DEFAULT_DEPLOYMENT environment variable to Local, Docker, Weaviate, or Custom.
First thing you need to do is to add your data. You can do this by clicking on Import Data and selecting either Add Files, Add Directory, or Add URL tab. Here you can add all your files that you want to ingest.
You can then configure every file individually by selecting the file and clicking on Overview or Configure tab.
With Data imported, you can use the Chat page to ask any related questions. You will receive relevant chunks that are semantically relevant to your question and an answer generated by your choosen model. You can configure the RAG pipeline under the Config tab.
Your contributions are always welcome! Feel free to contribute ideas, feedback, or create issues and bug reports if you find any! Before contributing, please read the Contribution Guide. Visit our Weaviate Community Forum if you need any help!
You can learn more about Verba's architecture and implementation in its technical documentation and frontend documentation. It's recommended to have a look at them before making any contributions.
- Weaviate Embeeded currently not working on Windows yet
- Will be fixed in future versions, until then please use the Docker or WCS Deployment
-
Can I use pre-existing data from my Weaviate instance?
- No, unfortunatley not. Verba requires the data to be in a specific format to work. And as of now, this is only possible by importing data through the Verba UI.
-
Is Verba Multi-Lingual?
- This depends on your choosen Embedding and Generation Model whether they support multi-lingual data.
-
Can I use my Ollama Server with the Verba Docker?
- Yes, you can! Make sure the URL is set to:
OLLAMA_URL=http://host.docker.internal:11434 - If you're running on Linux, you might need to get the IP Gateway of the Ollama server:
OLLAMA_URL="http://YOUR-IP-OF-OLLAMA:11434"
- Yes, you can! Make sure the URL is set to:
-
How to clear Weaviate Embedded Storage?
- You'll find the stored data here:
~/.local/share/weaviate
- You'll find the stored data here:
-
How can I specify the port?
- You can use the port and host flag
verba start --port 9000 --host 0.0.0.0
- You can use the port and host flag
-
Can multiple users use Verba at the same time? How about role based access?
- Verba is designed and optimized for single user usage only. There are no plans on supporting multiple users or role based access in the near future.
-
Does Verba offer a API endpoint to use externally?
- No, right now Verba does not offer any useful API endpoints to interact with the application. The current FastAPI setup is optimized for the internal communication between the frontend and backend. It is not recommended to use it as a API endpoint. There are plans to add user-friendly