Build a Private ChatGPT for Your Company With Open WebUI on a VPS

When team members use standard free or consumer AI chatbots, their prompts, proprietary algorithms, client names, and trade secrets are frequently retained and incorporated into model training pools. Several global technology giants and financial institutions have already faced embarrassing intellectual property leaks caused by staff pasting proprietary codebases into public AI portals.

By deploying Open WebUI on a private cloud VPS located in Indian datacenters, your organization achieves total data sovereignty. Open WebUI never sends telemetry to external vendors without explicit permission. All internal conversation histories, user authentication sessions, and uploaded documents stay securely locked within encrypted PostgreSQL or SQLite databases on your private server, fully complying with corporate audit guidelines.

Open WebUI is a feature-rich, self-hosted web interface designed to replicate the OpenAI ChatGPT user experience while granting complete infrastructure control. You can connect it to your preferred computational backend: 1. Hybrid Cloud API Gateway (Cost-Effective): Configure Open WebUI to connect to corporate OpenAI, Anthropic, or DeepSeek API keys. Your company signs a Zero Data Retention (ZDR) commercial agreement with the provider, and Open WebUI enforces role-based access control (RBAC), preventing individual employees from having direct access to raw API keys. 2. 100% Offline Local Model Backend (Air-Gapped Privacy): Pair Open WebUI directly with a local Ollama daemon running on the same Linux VPS. All token generation occurs on the server CPU or GPU. Not a single byte of data ever leaves your VPS network interface.

Deploying Open WebUI with Docker Compose takes under 10 minutes on a SkyRainCloud Linux VPS. Create a dedicated project directory and define the multi-container stack:

version: '3.8'
services:
  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    restart: always
    ports:
      - "127.0.0.1:8080:8080"
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
      - OPENAI_API_KEY=sk-proj-yourCorporateAPIKeyHere
      - WEBUI_SECRET_KEY=GenerateA32BitSecretStringForSessionSecurity
      - DEFAULT_MODELS=llama3.3:latest,gpt-4o
      - ENABLE_SIGNUP=false # Restricts registration to admin invites
    volumes:
      - open-webui-data:/app/backend/data
    depends_on:

ollama: image: ollama/ollama:latest container_name: ollama restart: always volumes: - ollama-models:/root/.ollama volumes: open-webui-data: ollama-models: ```

Run docker compose up -d, then configure Nginx with Certbot SSL to expose https://ai.yourcompany.in to your staff.

Once deployed, the first user to register automatically becomes the instance Administrator. From the Admin Panel, you can enforce strict enterprise security controls: - Disable Public Signups: Set user registration to invite-only or integrate OAuth2/SAML with Google Workspace or Microsoft Entra ID (Azure AD) for single sign-on. - Role Permissions: Restrict junior staff to specific cost-efficient models (e.g., Llama 3.3 or GPT-4o-mini) while granting senior developers access to top-tier reasoning engines. - Model Sandboxing: Create custom system prompts (System Personas) preloaded with company style guides, customer support rules, or legal review templates.

Open WebUI features a built-in Retrieval-Augmented Generation (RAG) engine. Human resource managers, support leads, and technical engineers can upload PDF employee handbooks, API technical manuals, internal policy documents, and financial spreadsheets directly into the Workspace.

Open WebUI automatically parses the files, chunks the text, computes embeddings using a local embedding model, and stores the vectors in its integrated vector database. When staff members ask questions like *"What is our corporate policy on maternity leave?"* or *"How do we issue a credit note in Tally?"*, the private bot cites the exact paragraph from internal documents without leaking data to the public internet.