Dev.to
6/18/2026

I Cancelled My $240/Year ChatGPT Subscription. 30 Days Later, My Laptop Knows Me Better Than GPT-4 Ever Did.
Short summary
A developer cancelled his $240/year ChatGPT subscription and built a private AI system using Ollama for inference, nomic-embed-text for embeddings, and Qdrant as a vector database—running entirely on his laptop with semantic search over personal documents. The post includes architecture details, real-world debugging challenges (Docker hangs, encoding bugs), and demonstrates that the local system answers questions better than cloud AI.
- •Built local AI system using Ollama, embeddings, and Qdrant instead of paying for ChatGPT
- •System achieves semantic search over personal documents without cloud data transfer
- •Includes technical walkthrough, code snippets, and debugging solutions for Windows/Docker
Generated with AI, which can make mistakes.
Is this a good recommendation for you?



