Private AI Chat Assistant — ChatGPT-level chat, on private infrastructure

A fully private, self-hosted AI assistant: a ChatGPT-style chat interface running open-source language models on dedicated infrastructure — no data sent to third parties, no per-message fees, available 24/7 behind HTTPS with login.

The problem

AI chat tools are transformative, but the mainstream options have two costs people underestimate: every conversation is sent to a third party, and usage-based pricing grows with adoption. For sensitive use cases — internal documents, business strategy, personal data — that trade-off is often unacceptable.

The solution

A self-hosted AI stack on a Linux VPS: Ollama serves open-source models (Llama 3.2, Mistral variants), and Open WebUI provides a polished, ChatGPT-like interface with conversation history and model switching. The whole stack runs in Docker containers on an isolated network, published through nginx with TLS, login required, and public signup disabled.

• Privacy by architecture: prompts and answers never leave the machine.
• Predictable cost: a fixed monthly server bill instead of per-token pricing.
• Hardened: HTTPS via Let's Encrypt, authentication, intrusion prevention (CrowdSec + fail2ban) on the host.

Stack

Results

• Production AI chat with zero third-party data exposure.
• Unlimited usage at a fixed, predictable cost.
• A foundation reused for other AI features — including the assistant on this very site.