Open WebUI + Ertas

Overview

Open WebUI (formerly Ollama WebUI) is a self-hosted, extensible web interface that provides a ChatGPT-like experience for locally running models. It connects to Ollama and OpenAI-compatible backends, offering a polished multi-user chat platform that organizations can deploy on their own infrastructure. Features include conversation history, model switching, document upload for retrieval-augmented generation (RAG), web search integration, image generation, custom model presets, and a robust admin panel for managing users and permissions.

For teams and organizations, Open WebUI solves the last-mile problem of making local AI accessible to everyone. Rather than requiring each user to install desktop applications or use CLI tools, Open WebUI provides a centralized web interface that any team member can access through their browser. With role-based access control, usage monitoring, and support for multiple concurrent models, it transforms a collection of local models into a managed AI service that rivals commercial offerings while keeping all data on-premises.

How Ertas Integrates

The integration between Ertas and Open WebUI flows through Ollama as the inference layer. After fine-tuning a model in Ertas Studio, export it in GGUF format and register it with Ollama using the provided Modelfile. Open WebUI automatically detects all models available in the connected Ollama instance, so your Ertas-trained model appears in the model selector immediately. Users can start conversations with the fine-tuned model, upload documents for RAG-enhanced responses, and compare outputs against other models — all through the familiar chat interface.

This three-layer architecture — Ertas for training, Ollama for serving, Open WebUI for the user interface — provides a complete private AI platform. A machine learning team iterates on model quality in Ertas Studio, an operations team manages the Ollama deployment, and end users interact with the models through Open WebUI without needing to understand the underlying infrastructure. The admin panel provides usage analytics and access controls, while the RAG pipeline lets users augment the fine-tuned model's knowledge with their own documents at query time.

Getting Started

1. 1

   Fine-tune in Ertas Studio

   Upload your JSONL training data and run a fine-tuning job on Ertas's managed cloud GPUs. Evaluate the model against your test set before exporting.

2. 2

   Export GGUF and register with Ollama

   Download the fine-tuned model in GGUF format from Ertas Studio. Use the provided Modelfile to register it with Ollama using 'ollama create my-model -f Modelfile'.

3. 3

   Deploy Open WebUI

   Run Open WebUI using Docker with a single command. Point it at your Ollama instance to automatically discover all available models including your Ertas-trained model.

4. 4

   Configure users and permissions

   Set up user accounts and role-based access controls in the Open WebUI admin panel. Restrict model access by role if needed — for example, limit expensive large models to specific teams.

5. 5

   Upload documents for RAG

   Users can upload PDFs, text files, and other documents directly in the chat interface. Open WebUI indexes them locally and uses retrieval-augmented generation to ground responses in your data.

6. 6

   Monitor usage and iterate

   Review conversation logs and usage analytics in the admin panel. Feed insights back into the next Ertas fine-tuning iteration to improve model quality for your team's specific needs.

Benefits

• Self-hosted ChatGPT-like experience accessible from any browser on the network
• Multi-user support with role-based access control and admin management
• Built-in RAG pipeline for document-grounded responses without external services
• Automatic model discovery from connected Ollama instances
• Web search integration for combining local model intelligence with live information
• Docker deployment with a single command for quick infrastructure setup