What is OpenWebUI?
OpenWebUI is an open-source web interface designed to interact with language models (LLM) like ChatGPT, Claude, or local models. This solution allows you to create a customized user interface for your artificial intelligence needs, facilitating access and use of these technologies.
Whether you are a business looking to integrate AI into your processes or a developer wanting to create a custom interface, OpenWebUI offers a flexible and powerful alternative to proprietary interfaces.
For a professional OpenWebUI configuration tailored to your specific needs, our AI development agency helps you implement custom solutions.
Use Cases
OpenWebUI is particularly suitable for several professional use cases:
Proprietary document databases: creating AI assistants capable of searching and analyzing internal company documents
Advanced semantic search: implementing intelligent search systems in vast knowledge bases
Enterprise AI assistants: deploying conversational interfaces powered by your private and secure data
Intelligent customer support: integrating RAG capabilities to improve automatic responses based on your documentation
RAG Settings: Optimizing Document Search
To achieve relevant results with OpenWebUI, several RAG (Retrieval-Augmented Generation) parameters must be adjusted according to your use case. Here are the main parameters to configure:
1. Chunking (document segmentation)
Chunking determines the size and overlap of text fragments extracted from your documents. Typical parameters vary according to content type:
Long technical documents: chunk size 400-600 characters, overlap 50-100 characters
Short content (FAQ, notices): chunk size 200-300 characters, overlap 20-30 characters
Scientific or legal documents: chunk size 600-800 characters, overlap 100-150 characters to preserve context
2. Embedding model
The choice of embedding model directly influences the quality of semantic search. Common options include:
OpenAI text-embedding-3-small (1536 dimensions): ideal for most cases, good performance/cost balance
OpenAI text-embedding-3-large (3072 dimensions): better accuracy for complex content, higher cost
Local models (Sentence-BERT, etc.): for sensitive data requiring offline processing
3. Hybrid search (BM25 + vectors)
The combination of textual (BM25) and vector search significantly improves result relevance:
Product/catalog search: BM25 weight 0.6-0.7 to favor exact term matches
Conceptual search: BM25 weight 0.3-0.4 to prioritize semantic similarity
General balance: BM25 weight 0.5 for optimal hybrid search
4. Reranker and Top K
The reranker refines initial results to improve accuracy:
Initial Top K: 3-5 documents to balance relevance and speed
Reranker Top K: 2-3 documents to focus the final context
Relevance threshold: 0.15-0.25 to filter out low-relevance results
5. Metadata and filters
Structuring metadata allows faster search and reduces noise:
Add fields like
collection,type,year,personUse filters in queries to limit search to specific subsets
Create an index/glossary for instant searches of exact matches
OpenWebUI Setup: Basic Steps
Once OpenWebUI is installed, here are the essential steps to configure your instance:
1. Access the interface
Access the web interface via your browser at http://localhost:3000 (or the configured port).
2. Create administrator account
On first connection, create an administrator account. This account gives you access to all configuration and management features of the instance.
3. Connect to AI models
Configure the connection to your artificial intelligence models:
OpenAI (GPT-3.5, GPT-4)
: add your OpenAI API key in settings
Anthropic Claude
: configure your Anthropic API key
Local models
: connect to Ollama or other local model servers
4. Configure vector databases
To enable RAG capabilities, configure a vector database:
Chroma
: simple solution to get started
Qdrant
: performant for large amounts of data
Weaviate
: cloud-native option with advanced features
Milvus
: for large-scale deployments
5. Important environment variables
Customize OpenWebUI via environment variables:
OPENAI_API_KEY: your OpenAI API key
WEBUI_SECRET_KEY: secret key for security (generate a strong one)
WEBUI_URL: base URL of your instance (important for webhooks and callbacks)
DATA_DIR: data storage directory (default:
/app/backend/data)
6. Security configuration
For a production environment:
Enable user authentication
Configure permissions and roles
Enable HTTPS via reverse proxy (Nginx, Traefik)
Configure rate limiting to prevent abuse
7. Document import
To use RAG capabilities:
Import your documents via the interface or API
Configure chunking parameters according to your content type
Verify that embeddings are properly generated
Test semantic search with a few queries