I'll never pay for AI again
Online AI tools are getting good fast. But the subscriptions stack up, and the free tiers are limited or locked to weaker models. After building a local coding assistant for VS Code, I realized I don't need paid plans like GitHub Copilot. I can run strong models on my own machine-and keep my data to myself.
Local AI is ready for real work
Open-source models have leveled up. Llama, Mistral, Gemma, and DeepSeek can handle writing, brainstorming, coding help, and more. There isn't one model that does everything, but picking the right model for a specific task gets you far.
For focused tasks, a well-chosen local model can match popular web tools like ChatGPT or Copilot. You can switch models in seconds, try new releases, and avoid rate limits and usage caps.
Hardware: what you actually need
You can run local AI on most modern machines. The bigger the model, the more memory it needs.
- 7B parameters: ~8GB RAM free
- 13B parameters: ~16GB RAM
- 33B parameters: ~32GB RAM
Keep plenty of disk space-model files can be large. Smaller models are faster and great for simple tasks; larger models handle more complexity. Pick based on the job, not hype.
Run local AI with LM Studio (or Ollama)
Tools like LM Studio and Ollama make setup simple. I prefer LM Studio for the clean GUI, but use what fits your workflow.
Quick-start with LM Studio:
- 1) Install from the official site and launch the app.
- 2) Let it fetch any drivers/updates.
- 3) Open the Discover tab (magnifying glass) and search for a model.
- 4) Click the green Download button.
- 5) Go to Chat and select your downloaded model from the dropdown.
Start chatting right away. Many models can read local files; LM Studio labels capabilities like image input or image generation so you know what each model supports.
Privacy and control, by default
Free web chatbots are useful, but they're still web tools. Some free tiers limit commercial use or train on your data. That's a deal-breaker for sensitive work like client code, contracts, or internal documents.
Local AI keeps everything on your machine. No data collection. No training on your inputs. No hidden logging. You get the freedom to use AI for confidential projects, personal writing, or anything you'd rather keep offline.
Practical picks to start
- Mistral 7B Instruct: fast, capable, lightweight.
- Gemma 3: strong general assistant for everyday tasks.
Test a few models, bookmark your favorites, and map each model to a task. Writing, email drafts, meeting notes, code hints, spreadsheet formulas-cover your daily work without a monthly bill.
Cut the subscriptions (on your terms)
You don't have to cancel everything at once. Keep the essentials while you test local models. As you build confidence, drop the subscriptions you don't need.
I stopped paying for AI because I didn't have to. With a free afternoon and the right tools, you can do the same-own your setup, keep your data, and get the work done.
If you want a structured path to pick the right tools and skills for your role, explore AI courses by job.
Your membership also unlocks: