I Stopped Paying for AI APIs — Here's What I Run Locally Instead

Six months ago I was spending $120/month between ChatGPT Plus, Claude Pro, and various API calls for my home automation experiments. Today I spend $0 — and get better results. Here's exactly how.

The Breaking Point

The trigger was a $47 invoice from OpenAI for API usage during a weekend project. I was running a simple n8n workflow that summarized RSS feeds — something that should cost pennies. That's when I decided to run everything locally.

The Stack

My setup runs on a single machine with an RTX 5070 and 47GB of RAM:

• Ollama — serves local LLMs via a clean REST API, identical interface to OpenAI
• LiteLLM — unified proxy that maps OpenAI-style calls to any local model
• Open WebUI — browser interface for model switching and chat history
• n8n — automation engine connecting everything together

The Models

For most tasks, llama3.1:8b handles 90% of what I used to pay GPT-4 for. For complex reasoning I run qwen3:8b or gemma3:12b. Zero API costs. Zero data leaving my network.

The Setup in 15 Minutes

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull a model
ollama pull llama3.1:8b

# Run Open WebUI via Docker
docker run -d -p 3000:8080 \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  -v open-webui:/app/backend/data \
  ghcr.io/open-webui/open-webui:latest

What You Give Up

Honesty: local 8B models are not GPT-4. They are slower for complex multi-step reasoning. But for summarization, classification, and automation tasks? They are indistinguishable — and they are private, offline-capable, and free forever.

The $120/month I saved pays for the GPU in about 18 months. After that it is pure margin.