Uploaded March 2026 | Updated September 2026, 3 weeks ago

---

I was spending $52/month on ChatGPT, ElevenLabs, and image generation APIs. So I moved everything local — LLMs, text-to-speech, image generation — all running
on hardware I own, for $0/month.

Here's the full stack:
- Ollama + Open WebUI (private ChatGPT replacement)
- Chatterbox TTS (beats ElevenLabs in blind tests — MIT licensed)
- ComfyUI (Stable Diffusion + Flux image generation)
- Tailscale (access everything from anywhere, no port forwarding)

Everything runs in Docker on a single GPU machine.

⏱️ Timestamps
00:00 The cost problem
00:36 The full stack overview
01:12 Hardware: it's all about VRAM
02:21 Ollama + Open WebUI setup
03:41 Chatterbox TTS setup
04:51 ComfyUI image generation
05:47 Tailscale remote access
06:44 The big picture
07:06 Final thoughts + real numbers

🔗 Resources
Docker Compose files → (link in pinned comment)
Ollama: ollama.com
Open WebUI: github.com/open-webui/open-webui
Chatterbox TTS: github.com/travisvn/chatterbox-tts
ComfyUI: github.com/comfyanonymous/ComfyUI
Tailscale: tailscale.com

💰 Hardware Recommendations
- 8GB VRAM → Small LLMs + SDXL (entry point)
- 12GB VRAM → Chatterbox + 14B models
- 16GB+ VRAM → Flux image generation
- 24GB VRAM → Everything runs simultaneously

This voiceover was generated by Chatterbox running on local hardware.

#ai #homelab #selfhosted #ollama #docker #privacy